1 /* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "byte-order.h"
21 #include <arpa/inet.h>
24 #include <sys/socket.h>
26 #include <openflow/openflow.h>
31 #include <sys/socket.h>
32 #include <sys/types.h>
36 #include "classifier.h"
41 #include "dynamic-string.h"
48 #include "mac-learning.h"
52 #include "ofp-print.h"
54 #include "ofproto/netflow.h"
55 #include "ofproto/ofproto.h"
56 #include "ovsdb-data.h"
58 #include "poll-loop.h"
62 #include "socket-util.h"
63 #include "stream-ssl.h"
66 #include "system-stats.h"
71 #include "vswitchd/vswitch-idl.h"
72 #include "xenserver.h"
74 #include "sflow_api.h"
76 VLOG_DEFINE_THIS_MODULE(bridge);
78 COVERAGE_DEFINE(bridge_flush);
79 COVERAGE_DEFINE(bridge_process_flow);
80 COVERAGE_DEFINE(bridge_process_cfm);
81 COVERAGE_DEFINE(bridge_process_lacp);
82 COVERAGE_DEFINE(bridge_reconfigure);
83 COVERAGE_DEFINE(bridge_lacp_update);
91 struct dst builtin[32];
96 static void dst_set_init(struct dst_set *);
97 static void dst_set_add(struct dst_set *, const struct dst *);
98 static void dst_set_free(struct dst_set *);
101 /* These members are always valid. */
102 struct list port_elem; /* Element in struct port's "ifaces" list. */
103 struct port *port; /* Containing port. */
104 char *name; /* Host network device name. */
105 tag_type tag; /* Tag associated with this interface. */
106 long long delay_expires; /* Time after which 'enabled' may change. */
108 /* These members are valid only after bridge_reconfigure() causes them to
110 struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
111 int dp_ifidx; /* Index within kernel datapath. */
112 struct netdev *netdev; /* Network device. */
113 bool enabled; /* May be chosen for flows? */
114 bool up; /* Is the interface up? */
115 const char *type; /* Usually same as cfg->type. */
116 const struct ovsrec_interface *cfg;
118 /* LACP information. */
119 uint16_t lacp_priority; /* LACP port priority. */
122 #define BOND_MASK 0xff
124 struct iface *iface; /* Assigned iface, or NULL if none. */
125 uint64_t tx_bytes; /* Count of bytes recently transmitted. */
126 tag_type tag; /* Tag for bond_entry<->iface association. */
130 BM_TCP, /* Transport Layer Load Balance. */
131 BM_SLB, /* Source Load Balance. */
132 BM_AB /* Active Backup. */
135 #define MAX_MIRRORS 32
136 typedef uint32_t mirror_mask_t;
137 #define MIRROR_MASK_C(X) UINT32_C(X)
138 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
140 struct bridge *bridge;
143 struct uuid uuid; /* UUID of this "mirror" record in database. */
145 /* Selection criteria. */
146 struct sset src_ports; /* Source port names. */
147 struct sset dst_ports; /* Destination port names. */
152 struct port *out_port;
156 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
158 struct bridge *bridge;
159 struct hmap_node hmap_node; /* Element in struct bridge's "ports" hmap. */
162 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
163 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
164 * NULL if all VLANs are trunked. */
165 const struct ovsrec_port *cfg;
168 struct netdev_monitor *monitor; /* Tracks carrier. NULL if miimon. */
169 long long int miimon_interval; /* Miimon status refresh interval. */
170 long long int miimon_next_update; /* Time of next miimon update. */
172 /* An ordinary bridge port has 1 interface.
173 * A bridge port for bonding has at least 2 interfaces. */
174 struct list ifaces; /* List of "struct iface"s. */
175 size_t n_ifaces; /* list_size(ifaces). */
178 enum bond_mode bond_mode; /* Type of the bond. BM_SLB is the default. */
179 struct iface *active_iface; /* iface on which bcasts accepted, or NULL. */
180 tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */
181 int updelay, downdelay; /* Delay before iface goes up/down, in ms. */
182 bool bond_fake_iface; /* Fake a bond interface for legacy compat? */
183 long long int bond_next_fake_iface_update; /* Time of next update. */
185 /* LACP information. */
186 struct lacp *lacp; /* LACP object. NULL if LACP is disabled. */
187 bool lacp_active; /* True if LACP is active */
188 bool lacp_fast; /* True if LACP is in fast mode. */
189 uint16_t lacp_priority; /* LACP system priority. */
191 /* SLB specific bonding info. */
192 struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */
193 int bond_rebalance_interval; /* Interval between rebalances, in ms. */
194 long long int bond_next_rebalance; /* Next rebalancing time. */
196 /* Port mirroring info. */
197 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
198 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
199 bool is_mirror_output_port; /* Does port mirroring send frames here? */
203 struct list node; /* Node in global list of bridges. */
204 char *name; /* User-specified arbitrary name. */
205 struct mac_learning *ml; /* MAC learning table. */
206 uint8_t ea[ETH_ADDR_LEN]; /* Bridge Ethernet Address. */
207 uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */
208 const struct ovsrec_bridge *cfg;
210 /* OpenFlow switch processing. */
211 struct ofproto *ofproto; /* OpenFlow switch. */
213 /* Kernel datapath information. */
214 struct dpif *dpif; /* Datapath. */
215 struct hmap ifaces; /* Contains "struct iface"s. */
218 struct hmap ports; /* "struct port"s indexed by name. */
219 struct shash iface_by_name; /* "struct iface"s indexed by name. */
222 bool has_bonded_ports;
227 /* Port mirroring. */
228 struct mirror *mirrors[MAX_MIRRORS];
231 /* List of all bridges. */
232 static struct list all_bridges = LIST_INITIALIZER(&all_bridges);
234 /* OVSDB IDL used to obtain configuration. */
235 static struct ovsdb_idl *idl;
237 /* Each time this timer expires, the bridge fetches systems and interface
238 * statistics and pushes them into the database. */
239 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
240 static long long int stats_timer = LLONG_MIN;
242 /* Stores the time after which CFM statistics may be written to the database.
243 * Only updated when changes to the database require rate limiting. */
244 #define CFM_LIMIT_INTERVAL (1 * 1000) /* In milliseconds. */
245 static long long int cfm_limiter = LLONG_MIN;
247 static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
248 static void bridge_destroy(struct bridge *);
249 static struct bridge *bridge_lookup(const char *name);
250 static unixctl_cb_func bridge_unixctl_dump_flows;
251 static unixctl_cb_func bridge_unixctl_reconnect;
252 static int bridge_run_one(struct bridge *);
253 static size_t bridge_get_controllers(const struct bridge *br,
254 struct ovsrec_controller ***controllersp);
255 static void bridge_reconfigure_one(struct bridge *);
256 static void bridge_reconfigure_remotes(struct bridge *,
257 const struct sockaddr_in *managers,
259 static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces);
260 static void bridge_fetch_dp_ifaces(struct bridge *);
261 static void bridge_flush(struct bridge *);
262 static void bridge_pick_local_hw_addr(struct bridge *,
263 uint8_t ea[ETH_ADDR_LEN],
264 struct iface **hw_addr_iface);
265 static uint64_t bridge_pick_datapath_id(struct bridge *,
266 const uint8_t bridge_ea[ETH_ADDR_LEN],
267 struct iface *hw_addr_iface);
268 static uint64_t dpid_from_hash(const void *, size_t nbytes);
270 static unixctl_cb_func bridge_unixctl_fdb_show;
271 static unixctl_cb_func cfm_unixctl_show;
272 static unixctl_cb_func qos_unixctl_show;
274 static void bond_init(void);
275 static void bond_run(struct port *);
276 static void bond_wait(struct port *);
277 static void bond_rebalance_port(struct port *);
278 static void bond_send_learning_packets(struct port *);
279 static void bond_enable_slave(struct iface *iface, bool enable);
281 static void port_run(struct port *);
282 static void port_wait(struct port *);
283 static struct port *port_create(struct bridge *, const char *name);
284 static void port_reconfigure(struct port *, const struct ovsrec_port *);
285 static void port_del_ifaces(struct port *, const struct ovsrec_port *);
286 static void port_destroy(struct port *);
287 static struct port *port_lookup(const struct bridge *, const char *name);
288 static struct iface *port_lookup_iface(const struct port *, const char *name);
289 static struct iface *port_get_an_iface(const struct port *);
290 static struct port *port_from_dp_ifidx(const struct bridge *,
292 static void port_update_bonding(struct port *);
293 static void port_update_lacp(struct port *);
295 static void mirror_create(struct bridge *, struct ovsrec_mirror *);
296 static void mirror_destroy(struct mirror *);
297 static void mirror_reconfigure(struct bridge *);
298 static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *);
299 static bool vlan_is_mirrored(const struct mirror *, int vlan);
301 static struct iface *iface_create(struct port *port,
302 const struct ovsrec_interface *if_cfg);
303 static void iface_destroy(struct iface *);
304 static struct iface *iface_lookup(const struct bridge *, const char *name);
305 static struct iface *iface_find(const char *name);
306 static struct iface *iface_from_dp_ifidx(const struct bridge *,
308 static void iface_set_mac(struct iface *);
309 static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport);
310 static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
311 static void iface_update_cfm(struct iface *);
312 static bool iface_refresh_cfm_stats(struct iface *iface);
313 static void iface_update_carrier(struct iface *);
314 static bool iface_get_carrier(const struct iface *);
316 static void shash_from_ovs_idl_map(char **keys, char **values, size_t n,
318 static void shash_to_ovs_idl_map(struct shash *,
319 char ***keys, char ***values, size_t *n);
321 /* Hooks into ofproto processing. */
322 static struct ofhooks bridge_ofhooks;
324 /* Public functions. */
326 /* Initializes the bridge module, configuring it to obtain its configuration
327 * from an OVSDB server accessed over 'remote', which should be a string in a
328 * form acceptable to ovsdb_idl_create(). */
330 bridge_init(const char *remote)
332 /* Create connection to database. */
333 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
335 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_cur_cfg);
336 ovsdb_idl_omit_alert(idl, &ovsrec_open_vswitch_col_statistics);
337 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids);
338 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_ovs_version);
339 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_db_version);
340 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_type);
341 ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_system_version);
343 ovsdb_idl_omit_alert(idl, &ovsrec_bridge_col_datapath_id);
344 ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids);
346 ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids);
347 ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge);
349 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_admin_state);
350 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_duplex);
351 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_speed);
352 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_link_state);
353 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_mtu);
354 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_ofport);
355 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_statistics);
356 ovsdb_idl_omit_alert(idl, &ovsrec_interface_col_status);
357 ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids);
359 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_is_connected);
360 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_role);
361 ovsdb_idl_omit_alert(idl, &ovsrec_controller_col_status);
362 ovsdb_idl_omit(idl, &ovsrec_controller_col_external_ids);
364 ovsdb_idl_omit_alert(idl, &ovsrec_maintenance_point_col_fault);
366 ovsdb_idl_omit_alert(idl, &ovsrec_monitor_col_fault);
368 ovsdb_idl_omit(idl, &ovsrec_qos_col_external_ids);
370 ovsdb_idl_omit(idl, &ovsrec_queue_col_external_ids);
372 ovsdb_idl_omit(idl, &ovsrec_mirror_col_external_ids);
374 ovsdb_idl_omit(idl, &ovsrec_netflow_col_external_ids);
376 ovsdb_idl_omit(idl, &ovsrec_sflow_col_external_ids);
378 ovsdb_idl_omit(idl, &ovsrec_manager_col_external_ids);
379 ovsdb_idl_omit(idl, &ovsrec_manager_col_inactivity_probe);
380 ovsdb_idl_omit(idl, &ovsrec_manager_col_is_connected);
381 ovsdb_idl_omit(idl, &ovsrec_manager_col_max_backoff);
382 ovsdb_idl_omit(idl, &ovsrec_manager_col_status);
384 ovsdb_idl_omit(idl, &ovsrec_ssl_col_external_ids);
386 /* Register unixctl commands. */
387 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL);
388 unixctl_command_register("cfm/show", cfm_unixctl_show, NULL);
389 unixctl_command_register("qos/show", qos_unixctl_show, NULL);
390 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows,
392 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect,
401 struct bridge *br, *next_br;
403 LIST_FOR_EACH_SAFE (br, next_br, node, &all_bridges) {
406 ovsdb_idl_destroy(idl);
409 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
410 * but for which the ovs-vswitchd configuration 'cfg' is required. */
412 bridge_configure_once(const struct ovsrec_open_vswitch *cfg)
414 static bool already_configured_once;
415 struct sset bridge_names;
416 struct sset dpif_names, dpif_types;
420 /* Only do this once per ovs-vswitchd run. */
421 if (already_configured_once) {
424 already_configured_once = true;
426 stats_timer = time_msec() + STATS_INTERVAL;
428 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
429 sset_init(&bridge_names);
430 for (i = 0; i < cfg->n_bridges; i++) {
431 sset_add(&bridge_names, cfg->bridges[i]->name);
434 /* Iterate over all system dpifs and delete any of them that do not appear
436 sset_init(&dpif_names);
437 sset_init(&dpif_types);
438 dp_enumerate_types(&dpif_types);
439 SSET_FOR_EACH (type, &dpif_types) {
442 dp_enumerate_names(type, &dpif_names);
444 /* Delete each dpif whose name is not in 'bridge_names'. */
445 SSET_FOR_EACH (name, &dpif_names) {
446 if (!sset_contains(&bridge_names, name)) {
450 retval = dpif_open(name, type, &dpif);
458 sset_destroy(&bridge_names);
459 sset_destroy(&dpif_names);
460 sset_destroy(&dpif_types);
463 /* Callback for iterate_and_prune_ifaces(). */
465 check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED)
467 if (!iface->netdev) {
468 /* We already reported a related error, don't bother duplicating it. */
472 if (iface->dp_ifidx < 0) {
473 VLOG_ERR("%s interface not in %s, dropping",
474 iface->name, dpif_name(br->dpif));
478 VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif),
479 iface->name, iface->dp_ifidx);
483 /* Callback for iterate_and_prune_ifaces(). */
485 set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface,
486 void *aux OVS_UNUSED)
488 /* Set policing attributes. */
489 netdev_set_policing(iface->netdev,
490 iface->cfg->ingress_policing_rate,
491 iface->cfg->ingress_policing_burst);
493 /* Set MAC address of internal interfaces other than the local
495 iface_set_mac(iface);
500 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
501 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
502 * deletes from 'br' any ports that no longer have any interfaces. */
504 iterate_and_prune_ifaces(struct bridge *br,
505 bool (*cb)(struct bridge *, struct iface *,
509 struct port *port, *next_port;
511 HMAP_FOR_EACH_SAFE (port, next_port, hmap_node, &br->ports) {
512 struct iface *iface, *next_iface;
514 LIST_FOR_EACH_SAFE (iface, next_iface, port_elem, &port->ifaces) {
515 if (!cb(br, iface, aux)) {
516 iface_set_ofport(iface->cfg, -1);
517 iface_destroy(iface);
521 if (!port->n_ifaces) {
522 VLOG_WARN("%s port has no interfaces, dropping", port->name);
528 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
529 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
530 * responsible for freeing '*managersp' (with free()).
532 * You may be asking yourself "why does ovs-vswitchd care?", because
533 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
534 * should not be and in fact is not directly involved in that. But
535 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
536 * it has to tell in-band control where the managers are to enable that.
537 * (Thus, only managers connected in-band are collected.)
540 collect_in_band_managers(const struct ovsrec_open_vswitch *ovs_cfg,
541 struct sockaddr_in **managersp, size_t *n_managersp)
543 struct sockaddr_in *managers = NULL;
544 size_t n_managers = 0;
548 /* Collect all of the potential targets from the "targets" columns of the
549 * rows pointed to by "manager_options", excluding any that are
552 for (i = 0; i < ovs_cfg->n_manager_options; i++) {
553 struct ovsrec_manager *m = ovs_cfg->manager_options[i];
555 if (m->connection_mode && !strcmp(m->connection_mode, "out-of-band")) {
556 sset_find_and_delete(&targets, m->target);
558 sset_add(&targets, m->target);
562 /* Now extract the targets' IP addresses. */
563 if (!sset_is_empty(&targets)) {
566 managers = xmalloc(sset_count(&targets) * sizeof *managers);
567 SSET_FOR_EACH (target, &targets) {
568 struct sockaddr_in *sin = &managers[n_managers];
570 if ((!strncmp(target, "tcp:", 4)
571 && inet_parse_active(target + 4, JSONRPC_TCP_PORT, sin)) ||
572 (!strncmp(target, "ssl:", 4)
573 && inet_parse_active(target + 4, JSONRPC_SSL_PORT, sin))) {
578 sset_destroy(&targets);
580 *managersp = managers;
581 *n_managersp = n_managers;
585 bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg)
587 struct shash old_br, new_br;
588 struct shash_node *node;
589 struct bridge *br, *next;
590 struct sockaddr_in *managers;
593 int sflow_bridge_number;
595 COVERAGE_INC(bridge_reconfigure);
597 collect_in_band_managers(ovs_cfg, &managers, &n_managers);
599 /* Collect old and new bridges. */
602 LIST_FOR_EACH (br, node, &all_bridges) {
603 shash_add(&old_br, br->name, br);
605 for (i = 0; i < ovs_cfg->n_bridges; i++) {
606 const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i];
607 if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) {
608 VLOG_WARN("more than one bridge named %s", br_cfg->name);
612 /* Get rid of deleted bridges and add new bridges. */
613 LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
614 struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
621 SHASH_FOR_EACH (node, &new_br) {
622 const char *br_name = node->name;
623 const struct ovsrec_bridge *br_cfg = node->data;
624 br = shash_find_data(&old_br, br_name);
626 /* If the bridge datapath type has changed, we need to tear it
627 * down and recreate. */
628 if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) {
630 bridge_create(br_cfg);
633 bridge_create(br_cfg);
636 shash_destroy(&old_br);
637 shash_destroy(&new_br);
639 /* Reconfigure all bridges. */
640 LIST_FOR_EACH (br, node, &all_bridges) {
641 bridge_reconfigure_one(br);
644 /* Add and delete ports on all datapaths.
646 * The kernel will reject any attempt to add a given port to a datapath if
647 * that port already belongs to a different datapath, so we must do all
648 * port deletions before any port additions. */
649 LIST_FOR_EACH (br, node, &all_bridges) {
650 struct dpif_port_dump dump;
651 struct shash want_ifaces;
652 struct dpif_port dpif_port;
654 bridge_get_all_ifaces(br, &want_ifaces);
655 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
656 if (!shash_find(&want_ifaces, dpif_port.name)
657 && strcmp(dpif_port.name, br->name)) {
658 int retval = dpif_port_del(br->dpif, dpif_port.port_no);
660 VLOG_WARN("failed to remove %s interface from %s: %s",
661 dpif_port.name, dpif_name(br->dpif),
666 shash_destroy(&want_ifaces);
668 LIST_FOR_EACH (br, node, &all_bridges) {
669 struct shash cur_ifaces, want_ifaces;
670 struct dpif_port_dump dump;
671 struct dpif_port dpif_port;
673 /* Get the set of interfaces currently in this datapath. */
674 shash_init(&cur_ifaces);
675 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
676 struct dpif_port *port_info = xmalloc(sizeof *port_info);
677 dpif_port_clone(port_info, &dpif_port);
678 shash_add(&cur_ifaces, dpif_port.name, port_info);
681 /* Get the set of interfaces we want on this datapath. */
682 bridge_get_all_ifaces(br, &want_ifaces);
684 hmap_clear(&br->ifaces);
685 SHASH_FOR_EACH (node, &want_ifaces) {
686 const char *if_name = node->name;
687 struct iface *iface = node->data;
688 struct dpif_port *dpif_port;
692 type = iface ? iface->type : "internal";
693 dpif_port = shash_find_data(&cur_ifaces, if_name);
695 /* If we have a port or a netdev already, and it's not the type we
696 * want, then delete the port (if any) and close the netdev (if
698 if ((dpif_port && strcmp(dpif_port->type, type))
699 || (iface && iface->netdev
700 && strcmp(type, netdev_get_type(iface->netdev)))) {
702 error = ofproto_port_del(br->ofproto, dpif_port->port_no);
709 netdev_close(iface->netdev);
710 iface->netdev = NULL;
714 /* If the port doesn't exist or we don't have the netdev open,
715 * we need to do more work. */
716 if (!dpif_port || (iface && !iface->netdev)) {
717 struct netdev_options options;
718 struct netdev *netdev;
721 /* First open the network device. */
722 options.name = if_name;
724 options.args = &args;
725 options.ethertype = NETDEV_ETH_TYPE_NONE;
729 shash_from_ovs_idl_map(iface->cfg->key_options,
730 iface->cfg->value_options,
731 iface->cfg->n_options, &args);
733 error = netdev_open(&options, &netdev);
734 shash_destroy(&args);
737 VLOG_WARN("could not open network device %s (%s)",
738 if_name, strerror(error));
742 /* Then add the port if we haven't already. */
744 error = dpif_port_add(br->dpif, netdev, NULL);
746 netdev_close(netdev);
747 if (error == EFBIG) {
748 VLOG_ERR("ran out of valid port numbers on %s",
749 dpif_name(br->dpif));
752 VLOG_WARN("failed to add %s interface to %s: %s",
753 if_name, dpif_name(br->dpif),
760 /* Update 'iface'. */
762 iface->netdev = netdev;
763 iface->enabled = iface_get_carrier(iface);
764 iface->up = iface->enabled;
766 } else if (iface && iface->netdev) {
770 shash_from_ovs_idl_map(iface->cfg->key_options,
771 iface->cfg->value_options,
772 iface->cfg->n_options, &args);
773 netdev_set_config(iface->netdev, &args);
774 shash_destroy(&args);
777 shash_destroy(&want_ifaces);
779 SHASH_FOR_EACH (node, &cur_ifaces) {
780 struct dpif_port *port_info = node->data;
781 dpif_port_destroy(port_info);
784 shash_destroy(&cur_ifaces);
786 sflow_bridge_number = 0;
787 LIST_FOR_EACH (br, node, &all_bridges) {
790 struct iface *local_iface;
791 struct iface *hw_addr_iface;
794 bridge_fetch_dp_ifaces(br);
796 iterate_and_prune_ifaces(br, check_iface, NULL);
798 /* Pick local port hardware address, datapath ID. */
799 bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
800 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
802 int error = netdev_set_etheraddr(local_iface->netdev, ea);
804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
805 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
806 "Ethernet address: %s",
807 br->name, strerror(error));
810 memcpy(br->ea, ea, ETH_ADDR_LEN);
812 dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface);
813 ofproto_set_datapath_id(br->ofproto, dpid);
815 dpid_string = xasprintf("%016"PRIx64, dpid);
816 ovsrec_bridge_set_datapath_id(br->cfg, dpid_string);
819 /* Set NetFlow configuration on this bridge. */
820 if (br->cfg->netflow) {
821 struct ovsrec_netflow *nf_cfg = br->cfg->netflow;
822 struct netflow_options opts;
824 memset(&opts, 0, sizeof opts);
826 dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
827 if (nf_cfg->engine_type) {
828 opts.engine_type = *nf_cfg->engine_type;
830 if (nf_cfg->engine_id) {
831 opts.engine_id = *nf_cfg->engine_id;
834 opts.active_timeout = nf_cfg->active_timeout;
835 if (!opts.active_timeout) {
836 opts.active_timeout = -1;
837 } else if (opts.active_timeout < 0) {
838 VLOG_WARN("bridge %s: active timeout interval set to negative "
839 "value, using default instead (%d seconds)", br->name,
840 NF_ACTIVE_TIMEOUT_DEFAULT);
841 opts.active_timeout = -1;
844 opts.add_id_to_iface = nf_cfg->add_id_to_interface;
845 if (opts.add_id_to_iface) {
846 if (opts.engine_id > 0x7f) {
847 VLOG_WARN("bridge %s: netflow port mangling may conflict "
848 "with another vswitch, choose an engine id less "
849 "than 128", br->name);
851 if (hmap_count(&br->ports) > 508) {
852 VLOG_WARN("bridge %s: netflow port mangling will conflict "
853 "with another port when more than 508 ports are "
858 sset_init(&opts.collectors);
859 sset_add_array(&opts.collectors,
860 nf_cfg->targets, nf_cfg->n_targets);
861 if (ofproto_set_netflow(br->ofproto, &opts)) {
862 VLOG_ERR("bridge %s: problem setting netflow collectors",
865 sset_destroy(&opts.collectors);
867 ofproto_set_netflow(br->ofproto, NULL);
870 /* Set sFlow configuration on this bridge. */
871 if (br->cfg->sflow) {
872 const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow;
873 struct ovsrec_controller **controllers;
874 struct ofproto_sflow_options oso;
875 size_t n_controllers;
877 memset(&oso, 0, sizeof oso);
879 sset_init(&oso.targets);
880 sset_add_array(&oso.targets,
881 sflow_cfg->targets, sflow_cfg->n_targets);
883 oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE;
884 if (sflow_cfg->sampling) {
885 oso.sampling_rate = *sflow_cfg->sampling;
888 oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL;
889 if (sflow_cfg->polling) {
890 oso.polling_interval = *sflow_cfg->polling;
893 oso.header_len = SFL_DEFAULT_HEADER_SIZE;
894 if (sflow_cfg->header) {
895 oso.header_len = *sflow_cfg->header;
898 oso.sub_id = sflow_bridge_number++;
899 oso.agent_device = sflow_cfg->agent;
901 oso.control_ip = NULL;
902 n_controllers = bridge_get_controllers(br, &controllers);
903 for (i = 0; i < n_controllers; i++) {
904 if (controllers[i]->local_ip) {
905 oso.control_ip = controllers[i]->local_ip;
909 ofproto_set_sflow(br->ofproto, &oso);
911 sset_destroy(&oso.targets);
913 ofproto_set_sflow(br->ofproto, NULL);
916 /* Update the controller and related settings. It would be more
917 * straightforward to call this from bridge_reconfigure_one(), but we
918 * can't do it there for two reasons. First, and most importantly, at
919 * that point we don't know the dp_ifidx of any interfaces that have
920 * been added to the bridge (because we haven't actually added them to
921 * the datapath). Second, at that point we haven't set the datapath ID
922 * yet; when a controller is configured, resetting the datapath ID will
923 * immediately disconnect from the controller, so it's better to set
924 * the datapath ID before the controller. */
925 bridge_reconfigure_remotes(br, managers, n_managers);
927 LIST_FOR_EACH (br, node, &all_bridges) {
930 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
934 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
935 netdev_monitor_add(port->monitor, iface->netdev);
938 port->miimon_next_update = 0;
941 port_update_lacp(port);
942 port_update_bonding(port);
944 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
945 iface_update_qos(iface, port->cfg->qos);
949 LIST_FOR_EACH (br, node, &all_bridges) {
950 iterate_and_prune_ifaces(br, set_iface_properties, NULL);
953 /* Some reconfiguration operations require the bridge to have been run at
955 LIST_FOR_EACH (br, node, &all_bridges) {
960 HMAP_FOR_EACH (iface, dp_ifidx_node, &br->ifaces) {
961 iface_update_cfm(iface);
967 /* ovs-vswitchd has completed initialization, so allow the process that
968 * forked us to exit successfully. */
969 daemonize_complete();
973 get_ovsrec_key_value(const struct ovsdb_idl_row *row,
974 const struct ovsdb_idl_column *column,
977 const struct ovsdb_datum *datum;
978 union ovsdb_atom atom;
981 datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING);
982 atom.string = (char *) key;
983 idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING);
984 return idx == UINT_MAX ? NULL : datum->values[idx].string;
988 bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key)
990 return get_ovsrec_key_value(&br_cfg->header_,
991 &ovsrec_bridge_col_other_config, key);
995 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
996 struct iface **hw_addr_iface)
1002 *hw_addr_iface = NULL;
1004 /* Did the user request a particular MAC? */
1005 hwaddr = bridge_get_other_config(br->cfg, "hwaddr");
1006 if (hwaddr && eth_addr_from_string(hwaddr, ea)) {
1007 if (eth_addr_is_multicast(ea)) {
1008 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
1009 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1010 } else if (eth_addr_is_zero(ea)) {
1011 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
1017 /* Otherwise choose the minimum non-local MAC address among all of the
1019 memset(ea, 0xff, ETH_ADDR_LEN);
1020 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1021 uint8_t iface_ea[ETH_ADDR_LEN];
1022 struct iface *candidate;
1023 struct iface *iface;
1025 /* Mirror output ports don't participate. */
1026 if (port->is_mirror_output_port) {
1030 /* Choose the MAC address to represent the port. */
1032 if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) {
1033 /* Find the interface with this Ethernet address (if any) so that
1034 * we can provide the correct devname to the caller. */
1035 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1036 uint8_t candidate_ea[ETH_ADDR_LEN];
1037 if (!netdev_get_etheraddr(candidate->netdev, candidate_ea)
1038 && eth_addr_equals(iface_ea, candidate_ea)) {
1043 /* Choose the interface whose MAC address will represent the port.
1044 * The Linux kernel bonding code always chooses the MAC address of
1045 * the first slave added to a bond, and the Fedora networking
1046 * scripts always add slaves to a bond in alphabetical order, so
1047 * for compatibility we choose the interface with the name that is
1048 * first in alphabetical order. */
1049 LIST_FOR_EACH (candidate, port_elem, &port->ifaces) {
1050 if (!iface || strcmp(candidate->name, iface->name) < 0) {
1055 /* The local port doesn't count (since we're trying to choose its
1056 * MAC address anyway). */
1057 if (iface->dp_ifidx == ODPP_LOCAL) {
1062 error = netdev_get_etheraddr(iface->netdev, iface_ea);
1064 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1065 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
1066 iface->name, strerror(error));
1071 /* Compare against our current choice. */
1072 if (!eth_addr_is_multicast(iface_ea) &&
1073 !eth_addr_is_local(iface_ea) &&
1074 !eth_addr_is_reserved(iface_ea) &&
1075 !eth_addr_is_zero(iface_ea) &&
1076 eth_addr_compare_3way(iface_ea, ea) < 0)
1078 memcpy(ea, iface_ea, ETH_ADDR_LEN);
1079 *hw_addr_iface = iface;
1082 if (eth_addr_is_multicast(ea)) {
1083 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
1084 *hw_addr_iface = NULL;
1085 VLOG_WARN("bridge %s: using default bridge Ethernet "
1086 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
1088 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
1089 br->name, ETH_ADDR_ARGS(ea));
1093 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1094 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1095 * an interface on 'br', then that interface must be passed in as
1096 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1097 * 'hw_addr_iface' must be passed in as a null pointer. */
1099 bridge_pick_datapath_id(struct bridge *br,
1100 const uint8_t bridge_ea[ETH_ADDR_LEN],
1101 struct iface *hw_addr_iface)
1104 * The procedure for choosing a bridge MAC address will, in the most
1105 * ordinary case, also choose a unique MAC that we can use as a datapath
1106 * ID. In some special cases, though, multiple bridges will end up with
1107 * the same MAC address. This is OK for the bridges, but it will confuse
1108 * the OpenFlow controller, because each datapath needs a unique datapath
1111 * Datapath IDs must be unique. It is also very desirable that they be
1112 * stable from one run to the next, so that policy set on a datapath
1115 const char *datapath_id;
1118 datapath_id = bridge_get_other_config(br->cfg, "datapath-id");
1119 if (datapath_id && dpid_from_string(datapath_id, &dpid)) {
1123 if (hw_addr_iface) {
1125 if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) {
1127 * A bridge whose MAC address is taken from a VLAN network device
1128 * (that is, a network device created with vconfig(8) or similar
1129 * tool) will have the same MAC address as a bridge on the VLAN
1130 * device's physical network device.
1132 * Handle this case by hashing the physical network device MAC
1133 * along with the VLAN identifier.
1135 uint8_t buf[ETH_ADDR_LEN + 2];
1136 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
1137 buf[ETH_ADDR_LEN] = vlan >> 8;
1138 buf[ETH_ADDR_LEN + 1] = vlan;
1139 return dpid_from_hash(buf, sizeof buf);
1142 * Assume that this bridge's MAC address is unique, since it
1143 * doesn't fit any of the cases we handle specially.
1148 * A purely internal bridge, that is, one that has no non-virtual
1149 * network devices on it at all, is more difficult because it has no
1150 * natural unique identifier at all.
1152 * When the host is a XenServer, we handle this case by hashing the
1153 * host's UUID with the name of the bridge. Names of bridges are
1154 * persistent across XenServer reboots, although they can be reused if
1155 * an internal network is destroyed and then a new one is later
1156 * created, so this is fairly effective.
1158 * When the host is not a XenServer, we punt by using a random MAC
1159 * address on each run.
1161 const char *host_uuid = xenserver_get_host_uuid();
1163 char *combined = xasprintf("%s,%s", host_uuid, br->name);
1164 dpid = dpid_from_hash(combined, strlen(combined));
1170 return eth_addr_to_uint64(bridge_ea);
1174 dpid_from_hash(const void *data, size_t n)
1176 uint8_t hash[SHA1_DIGEST_SIZE];
1178 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
1179 sha1_bytes(data, n, hash);
1180 eth_addr_mark_random(hash);
1181 return eth_addr_to_uint64(hash);
1185 iface_refresh_status(struct iface *iface)
1189 enum netdev_flags flags;
1198 if (!netdev_get_status(iface->netdev, &sh)) {
1200 char **keys, **values;
1202 shash_to_ovs_idl_map(&sh, &keys, &values, &n);
1203 ovsrec_interface_set_status(iface->cfg, keys, values, n);
1208 ovsrec_interface_set_status(iface->cfg, NULL, NULL, 0);
1211 shash_destroy_free_data(&sh);
1213 error = netdev_get_flags(iface->netdev, &flags);
1215 ovsrec_interface_set_admin_state(iface->cfg, flags & NETDEV_UP ? "up" : "down");
1218 ovsrec_interface_set_admin_state(iface->cfg, NULL);
1221 error = netdev_get_features(iface->netdev, ¤t, NULL, NULL, NULL);
1223 ovsrec_interface_set_duplex(iface->cfg,
1224 netdev_features_is_full_duplex(current)
1226 /* warning: uint64_t -> int64_t conversion */
1227 bps = netdev_features_to_bps(current);
1228 ovsrec_interface_set_link_speed(iface->cfg, &bps, 1);
1231 ovsrec_interface_set_duplex(iface->cfg, NULL);
1232 ovsrec_interface_set_link_speed(iface->cfg, NULL, 0);
1236 ovsrec_interface_set_link_state(iface->cfg,
1237 iface_get_carrier(iface) ? "up" : "down");
1239 error = netdev_get_mtu(iface->netdev, &mtu);
1240 if (!error && mtu != INT_MAX) {
1242 ovsrec_interface_set_mtu(iface->cfg, &mtu_64, 1);
1245 ovsrec_interface_set_mtu(iface->cfg, NULL, 0);
1249 /* Writes 'iface''s CFM statistics to the database. Returns true if anything
1250 * changed, false otherwise. */
1252 iface_refresh_cfm_stats(struct iface *iface)
1254 const struct ovsrec_monitor *mon;
1255 const struct cfm *cfm;
1256 bool changed = false;
1259 mon = iface->cfg->monitor;
1260 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1266 for (i = 0; i < mon->n_remote_mps; i++) {
1267 const struct ovsrec_maintenance_point *mp;
1268 const struct remote_mp *rmp;
1270 mp = mon->remote_mps[i];
1271 rmp = cfm_get_remote_mp(cfm, mp->mpid);
1273 if (mp->n_fault != 1 || mp->fault[0] != rmp->fault) {
1274 ovsrec_maintenance_point_set_fault(mp, &rmp->fault, 1);
1279 if (mon->n_fault != 1 || mon->fault[0] != cfm->fault) {
1280 ovsrec_monitor_set_fault(mon, &cfm->fault, 1);
1288 iface_refresh_stats(struct iface *iface)
1294 static const struct iface_stat iface_stats[] = {
1295 { "rx_packets", offsetof(struct netdev_stats, rx_packets) },
1296 { "tx_packets", offsetof(struct netdev_stats, tx_packets) },
1297 { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) },
1298 { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) },
1299 { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) },
1300 { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) },
1301 { "rx_errors", offsetof(struct netdev_stats, rx_errors) },
1302 { "tx_errors", offsetof(struct netdev_stats, tx_errors) },
1303 { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) },
1304 { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) },
1305 { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) },
1306 { "collisions", offsetof(struct netdev_stats, collisions) },
1308 enum { N_STATS = ARRAY_SIZE(iface_stats) };
1309 const struct iface_stat *s;
1311 char *keys[N_STATS];
1312 int64_t values[N_STATS];
1315 struct netdev_stats stats;
1317 /* Intentionally ignore return value, since errors will set 'stats' to
1318 * all-1s, and we will deal with that correctly below. */
1319 netdev_get_stats(iface->netdev, &stats);
1322 for (s = iface_stats; s < &iface_stats[N_STATS]; s++) {
1323 uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset);
1324 if (value != UINT64_MAX) {
1331 ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
1335 refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
1337 struct ovsdb_datum datum;
1341 get_system_stats(&stats);
1343 ovsdb_datum_from_shash(&datum, &stats);
1344 ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
1348 static inline const char *
1349 nx_role_to_str(enum nx_role role)
1354 case NX_ROLE_MASTER:
1359 return "*** INVALID ROLE ***";
1364 bridge_refresh_controller_status(const struct bridge *br)
1367 const struct ovsrec_controller *cfg;
1369 ofproto_get_ofproto_controller_info(br->ofproto, &info);
1371 OVSREC_CONTROLLER_FOR_EACH(cfg, idl) {
1372 struct ofproto_controller_info *cinfo =
1373 shash_find_data(&info, cfg->target);
1376 ovsrec_controller_set_is_connected(cfg, cinfo->is_connected);
1377 ovsrec_controller_set_role(cfg, nx_role_to_str(cinfo->role));
1378 ovsrec_controller_set_status(cfg, (char **) cinfo->pairs.keys,
1379 (char **) cinfo->pairs.values,
1382 ovsrec_controller_set_is_connected(cfg, false);
1383 ovsrec_controller_set_role(cfg, NULL);
1384 ovsrec_controller_set_status(cfg, NULL, NULL, 0);
1388 ofproto_free_ofproto_controller_info(&info);
1394 const struct ovsrec_open_vswitch *cfg;
1396 bool datapath_destroyed;
1397 bool database_changed;
1400 /* Let each bridge do the work that it needs to do. */
1401 datapath_destroyed = false;
1402 LIST_FOR_EACH (br, node, &all_bridges) {
1403 int error = bridge_run_one(br);
1405 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1406 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
1407 "forcing reconfiguration", br->name);
1408 datapath_destroyed = true;
1412 /* (Re)configure if necessary. */
1413 database_changed = ovsdb_idl_run(idl);
1414 cfg = ovsrec_open_vswitch_first(idl);
1416 /* Re-configure SSL. We do this on every trip through the main loop,
1417 * instead of just when the database changes, because the contents of the
1418 * key and certificate files can change without the database changing.
1420 * We do this before bridge_reconfigure() because that function might
1421 * initiate SSL connections and thus requires SSL to be configured. */
1422 if (cfg && cfg->ssl) {
1423 const struct ovsrec_ssl *ssl = cfg->ssl;
1425 stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate);
1426 stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert);
1429 if (database_changed || datapath_destroyed) {
1431 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
1433 bridge_configure_once(cfg);
1434 bridge_reconfigure(cfg);
1436 ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg);
1437 ovsdb_idl_txn_commit(txn);
1438 ovsdb_idl_txn_destroy(txn); /* XXX */
1440 /* We still need to reconfigure to avoid dangling pointers to
1441 * now-destroyed ovsrec structures inside bridge data. */
1442 static const struct ovsrec_open_vswitch null_cfg;
1444 bridge_reconfigure(&null_cfg);
1448 /* Refresh system and interface stats if necessary. */
1449 if (time_msec() >= stats_timer) {
1451 struct ovsdb_idl_txn *txn;
1453 txn = ovsdb_idl_txn_create(idl);
1454 LIST_FOR_EACH (br, node, &all_bridges) {
1457 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1458 struct iface *iface;
1460 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1461 iface_refresh_stats(iface);
1462 iface_refresh_status(iface);
1465 bridge_refresh_controller_status(br);
1467 refresh_system_stats(cfg);
1468 ovsdb_idl_txn_commit(txn);
1469 ovsdb_idl_txn_destroy(txn); /* XXX */
1472 stats_timer = time_msec() + STATS_INTERVAL;
1475 if (time_msec() >= cfm_limiter) {
1476 struct ovsdb_idl_txn *txn;
1477 bool changed = false;
1479 txn = ovsdb_idl_txn_create(idl);
1480 LIST_FOR_EACH (br, node, &all_bridges) {
1483 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1484 struct iface *iface;
1486 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
1487 changed = iface_refresh_cfm_stats(iface) || changed;
1493 cfm_limiter = time_msec() + CFM_LIMIT_INTERVAL;
1496 ovsdb_idl_txn_commit(txn);
1497 ovsdb_idl_txn_destroy(txn);
1506 LIST_FOR_EACH (br, node, &all_bridges) {
1509 ofproto_wait(br->ofproto);
1510 mac_learning_wait(br->ml);
1511 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1515 ovsdb_idl_wait(idl);
1516 poll_timer_wait_until(stats_timer);
1518 if (cfm_limiter > time_msec()) {
1519 poll_timer_wait_until(cfm_limiter);
1523 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1524 * configuration changes. */
1526 bridge_flush(struct bridge *br)
1528 COVERAGE_INC(bridge_flush);
1532 /* Bridge unixctl user interface functions. */
1534 bridge_unixctl_fdb_show(struct unixctl_conn *conn,
1535 const char *args, void *aux OVS_UNUSED)
1537 struct ds ds = DS_EMPTY_INITIALIZER;
1538 const struct bridge *br;
1539 const struct mac_entry *e;
1541 br = bridge_lookup(args);
1543 unixctl_command_reply(conn, 501, "no such bridge");
1547 ds_put_cstr(&ds, " port VLAN MAC Age\n");
1548 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
1549 struct port *port = e->port.p;
1550 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
1551 port_get_an_iface(port)->dp_ifidx,
1552 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
1554 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1558 /* CFM unixctl user interface functions. */
1560 cfm_unixctl_show(struct unixctl_conn *conn,
1561 const char *args, void *aux OVS_UNUSED)
1563 struct ds ds = DS_EMPTY_INITIALIZER;
1564 struct iface *iface;
1565 const struct cfm *cfm;
1567 iface = iface_find(args);
1569 unixctl_command_reply(conn, 501, "no such interface");
1573 cfm = ofproto_iface_get_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
1576 unixctl_command_reply(conn, 501, "CFM not enabled");
1580 cfm_dump_ds(cfm, &ds);
1581 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1585 /* QoS unixctl user interface functions. */
1587 struct qos_unixctl_show_cbdata {
1589 struct iface *iface;
1593 qos_unixctl_show_cb(unsigned int queue_id,
1594 const struct shash *details,
1597 struct qos_unixctl_show_cbdata *data = aux;
1598 struct ds *ds = data->ds;
1599 struct iface *iface = data->iface;
1600 struct netdev_queue_stats stats;
1601 struct shash_node *node;
1604 ds_put_cstr(ds, "\n");
1606 ds_put_format(ds, "Queue %u:\n", queue_id);
1608 ds_put_cstr(ds, "Default:\n");
1611 SHASH_FOR_EACH (node, details) {
1612 ds_put_format(ds, "\t%s: %s\n", node->name, (char *)node->data);
1615 error = netdev_get_queue_stats(iface->netdev, queue_id, &stats);
1617 if (stats.tx_packets != UINT64_MAX) {
1618 ds_put_format(ds, "\ttx_packets: %"PRIu64"\n", stats.tx_packets);
1621 if (stats.tx_bytes != UINT64_MAX) {
1622 ds_put_format(ds, "\ttx_bytes: %"PRIu64"\n", stats.tx_bytes);
1625 if (stats.tx_errors != UINT64_MAX) {
1626 ds_put_format(ds, "\ttx_errors: %"PRIu64"\n", stats.tx_errors);
1629 ds_put_format(ds, "\tFailed to get statistics for queue %u: %s",
1630 queue_id, strerror(error));
1635 qos_unixctl_show(struct unixctl_conn *conn,
1636 const char *args, void *aux OVS_UNUSED)
1638 struct ds ds = DS_EMPTY_INITIALIZER;
1639 struct shash sh = SHASH_INITIALIZER(&sh);
1640 struct iface *iface;
1642 struct shash_node *node;
1643 struct qos_unixctl_show_cbdata data;
1646 iface = iface_find(args);
1648 unixctl_command_reply(conn, 501, "no such interface");
1652 netdev_get_qos(iface->netdev, &type, &sh);
1654 if (*type != '\0') {
1655 ds_put_format(&ds, "QoS: %s %s\n", iface->name, type);
1657 SHASH_FOR_EACH (node, &sh) {
1658 ds_put_format(&ds, "%s: %s\n", node->name, (char *)node->data);
1663 error = netdev_dump_queues(iface->netdev, qos_unixctl_show_cb, &data);
1666 ds_put_format(&ds, "failed to dump queues: %s", strerror(error));
1668 unixctl_command_reply(conn, 200, ds_cstr(&ds));
1670 ds_put_format(&ds, "QoS not configured on %s\n", iface->name);
1671 unixctl_command_reply(conn, 501, ds_cstr(&ds));
1674 shash_destroy_free_data(&sh);
1678 /* Bridge reconfiguration functions. */
1679 static struct bridge *
1680 bridge_create(const struct ovsrec_bridge *br_cfg)
1685 assert(!bridge_lookup(br_cfg->name));
1686 br = xzalloc(sizeof *br);
1688 error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
1694 dpif_flow_flush(br->dpif);
1696 error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
1699 VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
1701 dpif_delete(br->dpif);
1702 dpif_close(br->dpif);
1707 br->name = xstrdup(br_cfg->name);
1709 br->ml = mac_learning_create();
1710 eth_addr_nicira_random(br->default_ea);
1712 hmap_init(&br->ports);
1713 hmap_init(&br->ifaces);
1714 shash_init(&br->iface_by_name);
1718 list_push_back(&all_bridges, &br->node);
1720 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
1726 bridge_destroy(struct bridge *br)
1729 struct port *port, *next;
1732 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1735 list_remove(&br->node);
1736 ofproto_destroy(br->ofproto);
1737 error = dpif_delete(br->dpif);
1738 if (error && error != ENOENT) {
1739 VLOG_ERR("failed to delete %s: %s",
1740 dpif_name(br->dpif), strerror(error));
1742 dpif_close(br->dpif);
1743 mac_learning_destroy(br->ml);
1744 hmap_destroy(&br->ifaces);
1745 hmap_destroy(&br->ports);
1746 shash_destroy(&br->iface_by_name);
1752 static struct bridge *
1753 bridge_lookup(const char *name)
1757 LIST_FOR_EACH (br, node, &all_bridges) {
1758 if (!strcmp(br->name, name)) {
1765 /* Handle requests for a listing of all flows known by the OpenFlow
1766 * stack, including those normally hidden. */
1768 bridge_unixctl_dump_flows(struct unixctl_conn *conn,
1769 const char *args, void *aux OVS_UNUSED)
1774 br = bridge_lookup(args);
1776 unixctl_command_reply(conn, 501, "Unknown bridge");
1781 ofproto_get_all_flows(br->ofproto, &results);
1783 unixctl_command_reply(conn, 200, ds_cstr(&results));
1784 ds_destroy(&results);
1787 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1788 * connections and reconnect. If BRIDGE is not specified, then all bridges
1789 * drop their controller connections and reconnect. */
1791 bridge_unixctl_reconnect(struct unixctl_conn *conn,
1792 const char *args, void *aux OVS_UNUSED)
1795 if (args[0] != '\0') {
1796 br = bridge_lookup(args);
1798 unixctl_command_reply(conn, 501, "Unknown bridge");
1801 ofproto_reconnect_controllers(br->ofproto);
1803 LIST_FOR_EACH (br, node, &all_bridges) {
1804 ofproto_reconnect_controllers(br->ofproto);
1807 unixctl_command_reply(conn, 200, NULL);
1811 bridge_run_one(struct bridge *br)
1816 error = ofproto_run1(br->ofproto);
1821 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
1823 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
1827 error = ofproto_run2(br->ofproto, br->flush);
1834 bridge_get_controllers(const struct bridge *br,
1835 struct ovsrec_controller ***controllersp)
1837 struct ovsrec_controller **controllers;
1838 size_t n_controllers;
1840 controllers = br->cfg->controller;
1841 n_controllers = br->cfg->n_controller;
1843 if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) {
1849 *controllersp = controllers;
1851 return n_controllers;
1855 bridge_reconfigure_one(struct bridge *br)
1857 enum ofproto_fail_mode fail_mode;
1858 struct port *port, *next;
1859 struct shash_node *node;
1860 struct shash new_ports;
1863 /* Collect new ports. */
1864 shash_init(&new_ports);
1865 for (i = 0; i < br->cfg->n_ports; i++) {
1866 const char *name = br->cfg->ports[i]->name;
1867 if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) {
1868 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1873 /* If we have a controller, then we need a local port. Complain if the
1874 * user didn't specify one.
1876 * XXX perhaps we should synthesize a port ourselves in this case. */
1877 if (bridge_get_controllers(br, NULL)) {
1878 char local_name[IF_NAMESIZE];
1881 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1882 local_name, sizeof local_name);
1883 if (!error && !shash_find(&new_ports, local_name)) {
1884 VLOG_WARN("bridge %s: controller specified but no local port "
1885 "(port named %s) defined",
1886 br->name, local_name);
1890 /* Get rid of deleted ports.
1891 * Get rid of deleted interfaces on ports that still exist. */
1892 HMAP_FOR_EACH_SAFE (port, next, hmap_node, &br->ports) {
1893 const struct ovsrec_port *port_cfg;
1895 port_cfg = shash_find_data(&new_ports, port->name);
1899 port_del_ifaces(port, port_cfg);
1903 /* Create new ports.
1904 * Add new interfaces to existing ports.
1905 * Reconfigure existing ports. */
1906 SHASH_FOR_EACH (node, &new_ports) {
1907 struct port *port = port_lookup(br, node->name);
1909 port = port_create(br, node->name);
1912 port_reconfigure(port, node->data);
1913 if (!port->n_ifaces) {
1914 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1915 br->name, port->name);
1919 shash_destroy(&new_ports);
1921 /* Set the fail-mode */
1922 fail_mode = !br->cfg->fail_mode
1923 || !strcmp(br->cfg->fail_mode, "standalone")
1924 ? OFPROTO_FAIL_STANDALONE
1925 : OFPROTO_FAIL_SECURE;
1926 if (ofproto_get_fail_mode(br->ofproto) != fail_mode
1927 && !ofproto_has_primary_controller(br->ofproto)) {
1928 ofproto_flush_flows(br->ofproto);
1930 ofproto_set_fail_mode(br->ofproto, fail_mode);
1932 /* Delete all flows if we're switching from connected to standalone or vice
1933 * versa. (XXX Should we delete all flows if we are switching from one
1934 * controller to another?) */
1936 /* Configure OpenFlow controller connection snooping. */
1937 if (!ofproto_has_snoops(br->ofproto)) {
1941 sset_add_and_free(&snoops, xasprintf("punix:%s/%s.snoop",
1942 ovs_rundir(), br->name));
1943 ofproto_set_snoops(br->ofproto, &snoops);
1944 sset_destroy(&snoops);
1947 mirror_reconfigure(br);
1950 /* Initializes 'oc' appropriately as a management service controller for
1953 * The caller must free oc->target when it is no longer needed. */
1955 bridge_ofproto_controller_for_mgmt(const struct bridge *br,
1956 struct ofproto_controller *oc)
1958 oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir(), br->name);
1959 oc->max_backoff = 0;
1960 oc->probe_interval = 60;
1961 oc->band = OFPROTO_OUT_OF_BAND;
1963 oc->burst_limit = 0;
1966 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1968 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c,
1969 struct ofproto_controller *oc)
1971 oc->target = c->target;
1972 oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8;
1973 oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5;
1974 oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band")
1975 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
1976 oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0;
1977 oc->burst_limit = (c->controller_burst_limit
1978 ? *c->controller_burst_limit : 0);
1981 /* Configures the IP stack for 'br''s local interface properly according to the
1982 * configuration in 'c'. */
1984 bridge_configure_local_iface_netdev(struct bridge *br,
1985 struct ovsrec_controller *c)
1987 struct netdev *netdev;
1988 struct in_addr mask, gateway;
1990 struct iface *local_iface;
1993 /* If there's no local interface or no IP address, give up. */
1994 local_iface = iface_from_dp_ifidx(br, ODPP_LOCAL);
1995 if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) {
1999 /* Bring up the local interface. */
2000 netdev = local_iface->netdev;
2001 netdev_turn_flags_on(netdev, NETDEV_UP, true);
2003 /* Configure the IP address and netmask. */
2004 if (!c->local_netmask
2005 || !inet_aton(c->local_netmask, &mask)
2007 mask.s_addr = guess_netmask(ip.s_addr);
2009 if (!netdev_set_in4(netdev, ip, mask)) {
2010 VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT,
2011 br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr));
2014 /* Configure the default gateway. */
2015 if (c->local_gateway
2016 && inet_aton(c->local_gateway, &gateway)
2017 && gateway.s_addr) {
2018 if (!netdev_add_router(netdev, gateway)) {
2019 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
2020 br->name, IP_ARGS(&gateway.s_addr));
2026 bridge_reconfigure_remotes(struct bridge *br,
2027 const struct sockaddr_in *managers,
2030 const char *disable_ib_str, *queue_id_str;
2031 bool disable_in_band = false;
2034 struct ovsrec_controller **controllers;
2035 size_t n_controllers;
2038 struct ofproto_controller *ocs;
2042 /* Check if we should disable in-band control on this bridge. */
2043 disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band");
2044 if (disable_ib_str && !strcmp(disable_ib_str, "true")) {
2045 disable_in_band = true;
2048 /* Set OpenFlow queue ID for in-band control. */
2049 queue_id_str = bridge_get_other_config(br->cfg, "in-band-queue");
2050 queue_id = queue_id_str ? strtol(queue_id_str, NULL, 10) : -1;
2051 ofproto_set_in_band_queue(br->ofproto, queue_id);
2053 if (disable_in_band) {
2054 ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0);
2056 ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers);
2058 had_primary = ofproto_has_primary_controller(br->ofproto);
2060 n_controllers = bridge_get_controllers(br, &controllers);
2062 ocs = xmalloc((n_controllers + 1) * sizeof *ocs);
2065 bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]);
2066 for (i = 0; i < n_controllers; i++) {
2067 struct ovsrec_controller *c = controllers[i];
2069 if (!strncmp(c->target, "punix:", 6)
2070 || !strncmp(c->target, "unix:", 5)) {
2071 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2073 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
2074 * domain sockets and overwriting arbitrary local files. */
2075 VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
2076 "\"%s\" due to possibility for remote exploit",
2077 dpif_name(br->dpif), c->target);
2081 bridge_configure_local_iface_netdev(br, c);
2082 bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]);
2083 if (disable_in_band) {
2084 ocs[n_ocs].band = OFPROTO_OUT_OF_BAND;
2089 ofproto_set_controllers(br->ofproto, ocs, n_ocs);
2090 free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */
2093 if (had_primary != ofproto_has_primary_controller(br->ofproto)) {
2094 ofproto_flush_flows(br->ofproto);
2097 /* If there are no controllers and the bridge is in standalone
2098 * mode, set up a flow that matches every packet and directs
2099 * them to OFPP_NORMAL (which goes to us). Otherwise, the
2100 * switch is in secure mode and we won't pass any traffic until
2101 * a controller has been defined and it tells us to do so. */
2103 && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) {
2104 union ofp_action action;
2105 struct cls_rule rule;
2107 memset(&action, 0, sizeof action);
2108 action.type = htons(OFPAT_OUTPUT);
2109 action.output.len = htons(sizeof action);
2110 action.output.port = htons(OFPP_NORMAL);
2111 cls_rule_init_catchall(&rule, 0);
2112 ofproto_add_flow(br->ofproto, &rule, &action, 1);
2117 bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces)
2122 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2123 struct iface *iface;
2125 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2126 shash_add_once(ifaces, iface->name, iface);
2128 if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) {
2129 shash_add_once(ifaces, port->name, NULL);
2134 /* For robustness, in case the administrator moves around datapath ports behind
2135 * our back, we re-check all the datapath port numbers here.
2137 * This function will set the 'dp_ifidx' members of interfaces that have
2138 * disappeared to -1, so only call this function from a context where those
2139 * 'struct iface's will be removed from the bridge. Otherwise, the -1
2140 * 'dp_ifidx'es will cause trouble later when we try to send them to the
2141 * datapath, which doesn't support UINT16_MAX+1 ports. */
2143 bridge_fetch_dp_ifaces(struct bridge *br)
2145 struct dpif_port_dump dump;
2146 struct dpif_port dpif_port;
2149 /* Reset all interface numbers. */
2150 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2151 struct iface *iface;
2153 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2154 iface->dp_ifidx = -1;
2157 hmap_clear(&br->ifaces);
2159 DPIF_PORT_FOR_EACH (&dpif_port, &dump, br->dpif) {
2160 struct iface *iface = iface_lookup(br, dpif_port.name);
2162 if (iface->dp_ifidx >= 0) {
2163 VLOG_WARN("%s reported interface %s twice",
2164 dpif_name(br->dpif), dpif_port.name);
2165 } else if (iface_from_dp_ifidx(br, dpif_port.port_no)) {
2166 VLOG_WARN("%s reported interface %"PRIu16" twice",
2167 dpif_name(br->dpif), dpif_port.port_no);
2169 iface->dp_ifidx = dpif_port.port_no;
2170 hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
2171 hash_int(iface->dp_ifidx, 0));
2174 iface_set_ofport(iface->cfg,
2175 (iface->dp_ifidx >= 0
2176 ? odp_port_to_ofp_port(iface->dp_ifidx)
2182 /* Bridge packet processing functions. */
2185 bond_is_tcp_hash(const struct port *port)
2187 return port->bond_mode == BM_TCP && lacp_negotiated(port->lacp);
2191 bond_hash_src(const uint8_t mac[ETH_ADDR_LEN], uint16_t vlan)
2193 return hash_bytes(mac, ETH_ADDR_LEN, vlan) & BOND_MASK;
2196 static int bond_hash_tcp(const struct flow *flow, uint16_t vlan)
2198 struct flow hash_flow;
2200 memcpy(&hash_flow, flow, sizeof hash_flow);
2201 hash_flow.vlan_tci = 0;
2203 /* The symmetric quality of this hash function is not required, but
2204 * flow_hash_symmetric_l4 already exists, and is sufficient for our
2205 * purposes, so we use it out of convenience. */
2206 return flow_hash_symmetric_l4(&hash_flow, vlan) & BOND_MASK;
2209 static struct bond_entry *
2210 lookup_bond_entry(const struct port *port, const struct flow *flow,
2213 assert(port->bond_mode != BM_AB);
2215 if (bond_is_tcp_hash(port)) {
2216 return &port->bond_hash[bond_hash_tcp(flow, vlan)];
2218 return &port->bond_hash[bond_hash_src(flow->dl_src, vlan)];
2222 static struct iface *
2223 bond_choose_iface(const struct port *port)
2225 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2226 struct iface *best_down_slave;
2227 struct iface *iface;
2229 best_down_slave = NULL;
2230 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2231 if (iface->enabled) {
2233 } else if ((!best_down_slave
2234 || iface->delay_expires < best_down_slave->delay_expires)
2235 && lacp_slave_may_enable(port->lacp, iface)) {
2236 best_down_slave = iface;
2240 if (best_down_slave) {
2241 VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay "
2242 "since no other interface is up",
2243 best_down_slave->name,
2244 best_down_slave->delay_expires - time_msec());
2245 bond_enable_slave(best_down_slave, true);
2248 return best_down_slave;
2252 choose_output_iface(const struct port *port, const struct flow *flow,
2253 uint16_t vlan, uint16_t *dp_ifidx, tag_type *tags)
2255 struct iface *iface;
2257 assert(port->n_ifaces);
2258 if (port->n_ifaces == 1) {
2259 iface = port_get_an_iface(port);
2260 } else if (port->bond_mode == BM_AB) {
2261 iface = port->active_iface;
2263 *tags |= port->no_ifaces_tag;
2267 struct bond_entry *e = lookup_bond_entry(port, flow, vlan);
2268 if (!e->iface || !e->iface->enabled) {
2269 /* XXX select interface properly. The current interface selection
2270 * is only good for testing the rebalancing code. */
2271 e->iface = bond_choose_iface(port);
2273 *tags |= port->no_ifaces_tag;
2276 e->tag = tag_create_random();
2281 *dp_ifidx = iface->dp_ifidx;
2282 *tags |= iface->tag; /* Currently only used for bonding. */
2287 bond_link_status_update(struct iface *iface)
2289 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2290 struct port *port = iface->port;
2291 bool up = iface->up && lacp_slave_may_enable(port->lacp, iface);
2292 int updelay, downdelay;
2294 updelay = port->updelay;
2295 downdelay = port->downdelay;
2297 if (lacp_negotiated(port->lacp)) {
2302 if ((up == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
2303 /* Nothing to do. */
2306 VLOG_INFO_RL(&rl, "interface %s: link state %s",
2307 iface->name, up ? "up" : "down");
2308 if (up == iface->enabled) {
2309 iface->delay_expires = LLONG_MAX;
2310 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
2311 iface->name, up ? "disabled" : "enabled");
2312 } else if (up && !port->active_iface) {
2313 bond_enable_slave(iface, true);
2315 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
2316 "other interface is up", iface->name, updelay);
2319 int delay = up ? updelay : downdelay;
2320 iface->delay_expires = time_msec() + delay;
2323 "interface %s: will be %s if it stays %s for %d ms",
2325 up ? "enabled" : "disabled",
2333 bond_choose_active_iface(struct port *port)
2335 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
2337 port->active_iface = bond_choose_iface(port);
2338 if (port->active_iface) {
2339 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
2340 port->name, port->active_iface->name);
2342 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
2348 bond_enable_slave(struct iface *iface, bool enable)
2350 struct port *port = iface->port;
2351 struct bridge *br = port->bridge;
2353 /* This acts as a recursion check. If the act of disabling a slave
2354 * causes a different slave to be enabled, the flag will allow us to
2355 * skip redundant work when we reenter this function. It must be
2356 * cleared on exit to keep things safe with multiple bonds. */
2357 static bool moving_active_iface = false;
2359 iface->delay_expires = LLONG_MAX;
2360 if (enable == iface->enabled) {
2364 iface->enabled = enable;
2365 if (!iface->enabled) {
2366 VLOG_WARN("interface %s: disabled", iface->name);
2367 ofproto_revalidate(br->ofproto, iface->tag);
2368 if (iface == port->active_iface) {
2369 /* Disabling a slave can lead to another slave being immediately
2370 * enabled if there will be no active slaves but one is waiting
2371 * on an updelay. In this case we do not need to run most of the
2372 * code for the newly enabled slave since there was no period
2373 * without an active slave and it is redundant with the disabling
2375 moving_active_iface = true;
2376 bond_choose_active_iface(port);
2378 bond_send_learning_packets(port);
2380 VLOG_WARN("interface %s: enabled", iface->name);
2381 if (!port->active_iface && !moving_active_iface) {
2382 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
2383 bond_choose_active_iface(port);
2384 bond_send_learning_packets(port);
2386 iface->tag = tag_create_random();
2389 moving_active_iface = false;
2392 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2393 * bond interface. */
2395 bond_update_fake_iface_stats(struct port *port)
2397 struct netdev_stats bond_stats;
2398 struct netdev *bond_dev;
2399 struct iface *iface;
2401 memset(&bond_stats, 0, sizeof bond_stats);
2403 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2404 struct netdev_stats slave_stats;
2406 if (!netdev_get_stats(iface->netdev, &slave_stats)) {
2407 /* XXX: We swap the stats here because they are swapped back when
2408 * reported by the internal device. The reason for this is
2409 * internal devices normally represent packets going into the system
2410 * but when used as fake bond device they represent packets leaving
2411 * the system. We really should do this in the internal device
2412 * itself because changing it here reverses the counts from the
2413 * perspective of the switch. However, the internal device doesn't
2414 * know what type of device it represents so we have to do it here
2416 bond_stats.tx_packets += slave_stats.rx_packets;
2417 bond_stats.tx_bytes += slave_stats.rx_bytes;
2418 bond_stats.rx_packets += slave_stats.tx_packets;
2419 bond_stats.rx_bytes += slave_stats.tx_bytes;
2423 if (!netdev_open_default(port->name, &bond_dev)) {
2424 netdev_set_stats(bond_dev, &bond_stats);
2425 netdev_close(bond_dev);
2430 bond_run(struct port *port)
2432 struct iface *iface;
2434 if (port->n_ifaces < 2) {
2438 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2439 bond_link_status_update(iface);
2442 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2443 if (time_msec() >= iface->delay_expires) {
2444 bond_enable_slave(iface, !iface->enabled);
2448 if (port->bond_fake_iface
2449 && time_msec() >= port->bond_next_fake_iface_update) {
2450 bond_update_fake_iface_stats(port);
2451 port->bond_next_fake_iface_update = time_msec() + 1000;
2456 bond_wait(struct port *port)
2458 struct iface *iface;
2460 if (port->n_ifaces < 2) {
2464 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2465 if (iface->delay_expires != LLONG_MAX) {
2466 poll_timer_wait_until(iface->delay_expires);
2470 if (port->bond_fake_iface) {
2471 poll_timer_wait_until(port->bond_next_fake_iface_update);
2476 set_dst(struct dst *dst, const struct flow *flow,
2477 const struct port *in_port, const struct port *out_port,
2480 dst->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
2481 : in_port->vlan >= 0 ? in_port->vlan
2482 : flow->vlan_tci == 0 ? OFP_VLAN_NONE
2483 : vlan_tci_to_vid(flow->vlan_tci));
2484 return choose_output_iface(out_port, flow, dst->vlan,
2485 &dst->dp_ifidx, tags);
2489 swap_dst(struct dst *p, struct dst *q)
2491 struct dst tmp = *p;
2496 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2497 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2498 * that we push to the datapath. We could in fact fully sort the array by
2499 * vlan, but in most cases there are at most two different vlan tags so that's
2500 * possibly overkill.) */
2502 partition_dsts(struct dst_set *set, int vlan)
2504 struct dst *first = set->dsts;
2505 struct dst *last = set->dsts + set->n;
2507 while (first != last) {
2509 * - All dsts < first have vlan == 'vlan'.
2510 * - All dsts >= last have vlan != 'vlan'.
2511 * - first < last. */
2512 while (first->vlan == vlan) {
2513 if (++first == last) {
2518 /* Same invariants, plus one additional:
2519 * - first->vlan != vlan.
2521 while (last[-1].vlan != vlan) {
2522 if (--last == first) {
2527 /* Same invariants, plus one additional:
2528 * - last[-1].vlan == vlan.*/
2529 swap_dst(first++, --last);
2534 mirror_mask_ffs(mirror_mask_t mask)
2536 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
2541 dst_set_init(struct dst_set *set)
2543 set->dsts = set->builtin;
2545 set->allocated = ARRAY_SIZE(set->builtin);
2549 dst_set_add(struct dst_set *set, const struct dst *dst)
2551 if (set->n >= set->allocated) {
2552 size_t new_allocated;
2553 struct dst *new_dsts;
2555 new_allocated = set->allocated * 2;
2556 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
2557 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
2561 set->dsts = new_dsts;
2562 set->allocated = new_allocated;
2564 set->dsts[set->n++] = *dst;
2568 dst_set_free(struct dst_set *set)
2570 if (set->dsts != set->builtin) {
2576 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
2579 for (i = 0; i < set->n; i++) {
2580 if (set->dsts[i].vlan == test->vlan
2581 && set->dsts[i].dp_ifidx == test->dp_ifidx) {
2589 port_trunks_vlan(const struct port *port, uint16_t vlan)
2591 return (port->vlan < 0
2592 && (!port->trunks || bitmap_is_set(port->trunks, vlan)));
2596 port_includes_vlan(const struct port *port, uint16_t vlan)
2598 return vlan == port->vlan || port_trunks_vlan(port, vlan);
2602 port_is_floodable(const struct port *port)
2604 struct iface *iface;
2606 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
2607 if (!ofproto_port_is_floodable(port->bridge->ofproto,
2615 /* Returns the tag for 'port''s active iface, or 'port''s no_ifaces_tag if
2616 * there is no active iface. */
2618 port_get_active_iface_tag(const struct port *port)
2620 return (port->active_iface
2621 ? port->active_iface->tag
2622 : port->no_ifaces_tag);
2625 /* Returns an arbitrary interface within 'port'.
2627 * 'port' must have at least one interface. */
2628 static struct iface *
2629 port_get_an_iface(const struct port *port)
2631 return CONTAINER_OF(list_front(&port->ifaces), struct iface, port_elem);
2635 compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan,
2636 const struct port *in_port, const struct port *out_port,
2637 struct dst_set *set, tag_type *tags, uint16_t *nf_output_iface)
2639 mirror_mask_t mirrors = in_port->src_mirrors;
2643 flow_vlan = vlan_tci_to_vid(flow->vlan_tci);
2644 if (flow_vlan == 0) {
2645 flow_vlan = OFP_VLAN_NONE;
2648 if (out_port == FLOOD_PORT) {
2651 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2653 && port_is_floodable(port)
2654 && port_includes_vlan(port, vlan)
2655 && !port->is_mirror_output_port
2656 && set_dst(&dst, flow, in_port, port, tags)) {
2657 mirrors |= port->dst_mirrors;
2658 dst_set_add(set, &dst);
2661 *nf_output_iface = NF_OUT_FLOOD;
2662 } else if (out_port && set_dst(&dst, flow, in_port, out_port, tags)) {
2663 dst_set_add(set, &dst);
2664 *nf_output_iface = dst.dp_ifidx;
2665 mirrors |= out_port->dst_mirrors;
2669 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
2670 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
2672 if (set_dst(&dst, flow, in_port, m->out_port, tags)
2673 && !dst_is_duplicate(set, &dst)) {
2674 dst_set_add(set, &dst);
2679 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
2680 if (port_includes_vlan(port, m->out_vlan)
2681 && set_dst(&dst, flow, in_port, port, tags))
2683 if (port->vlan < 0) {
2684 dst.vlan = m->out_vlan;
2686 if (dst_is_duplicate(set, &dst)) {
2690 /* Use the vlan tag on the original flow instead of
2691 * the one passed in the vlan parameter. This ensures
2692 * that we compare the vlan from before any implicit
2693 * tagging tags place. This is necessary because
2694 * dst->vlan is the final vlan, after removing implicit
2696 if (port == in_port && dst.vlan == flow_vlan) {
2697 /* Don't send out input port on same VLAN. */
2700 dst_set_add(set, &dst);
2705 mirrors &= mirrors - 1;
2708 partition_dsts(set, flow_vlan);
2711 static void OVS_UNUSED
2712 print_dsts(const struct dst_set *set)
2716 for (i = 0; i < set->n; i++) {
2717 const struct dst *dst = &set->dsts[i];
2719 printf(">p%"PRIu16, dst->dp_ifidx);
2720 if (dst->vlan != OFP_VLAN_NONE) {
2721 printf("v%"PRIu16, dst->vlan);
2727 compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan,
2728 const struct port *in_port, const struct port *out_port,
2729 tag_type *tags, struct ofpbuf *actions,
2730 uint16_t *nf_output_iface)
2737 compose_dsts(br, flow, vlan, in_port, out_port, &set, tags,
2740 cur_vlan = vlan_tci_to_vid(flow->vlan_tci);
2741 if (cur_vlan == 0) {
2742 cur_vlan = OFP_VLAN_NONE;
2744 for (i = 0; i < set.n; i++) {
2745 const struct dst *dst = &set.dsts[i];
2746 if (dst->vlan != cur_vlan) {
2747 if (dst->vlan == OFP_VLAN_NONE) {
2748 nl_msg_put_flag(actions, ODP_ACTION_ATTR_STRIP_VLAN);
2751 tci = htons(dst->vlan & VLAN_VID_MASK);
2752 tci |= flow->vlan_tci & htons(VLAN_PCP_MASK);
2753 nl_msg_put_be16(actions, ODP_ACTION_ATTR_SET_DL_TCI, tci);
2755 cur_vlan = dst->vlan;
2757 nl_msg_put_u32(actions, ODP_ACTION_ATTR_OUTPUT, dst->dp_ifidx);
2762 /* Returns the effective vlan of a packet, taking into account both the
2763 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2764 * the packet is untagged and -1 indicates it has an invalid header and
2765 * should be dropped. */
2766 static int flow_get_vlan(struct bridge *br, const struct flow *flow,
2767 struct port *in_port, bool have_packet)
2769 int vlan = vlan_tci_to_vid(flow->vlan_tci);
2770 if (in_port->vlan >= 0) {
2772 /* XXX support double tagging? */
2774 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2775 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2776 "packet received on port %s configured with "
2777 "implicit VLAN %"PRIu16,
2778 br->name, vlan, in_port->name, in_port->vlan);
2782 vlan = in_port->vlan;
2784 if (!port_includes_vlan(in_port, vlan)) {
2786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2787 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
2788 "packet received on port %s not configured for "
2790 br->name, vlan, in_port->name, vlan);
2799 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2800 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2801 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2803 is_gratuitous_arp(const struct flow *flow)
2805 return (flow->dl_type == htons(ETH_TYPE_ARP)
2806 && eth_addr_is_broadcast(flow->dl_dst)
2807 && (flow->nw_proto == ARP_OP_REPLY
2808 || (flow->nw_proto == ARP_OP_REQUEST
2809 && flow->nw_src == flow->nw_dst)));
2813 update_learning_table(struct bridge *br, const struct flow *flow, int vlan,
2814 struct port *in_port)
2816 struct mac_entry *mac;
2818 if (!mac_learning_may_learn(br->ml, flow->dl_src, vlan)) {
2822 mac = mac_learning_insert(br->ml, flow->dl_src, vlan);
2823 if (is_gratuitous_arp(flow)) {
2824 /* We don't want to learn from gratuitous ARP packets that are
2825 * reflected back over bond slaves so we lock the learning table. */
2826 if (in_port->n_ifaces == 1) {
2827 mac_entry_set_grat_arp_lock(mac);
2828 } else if (mac_entry_is_grat_arp_locked(mac)) {
2833 if (mac_entry_is_new(mac) || mac->port.p != in_port) {
2834 /* The log messages here could actually be useful in debugging,
2835 * so keep the rate limit relatively high. */
2836 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2837 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2838 "on port %s in VLAN %d",
2839 br->name, ETH_ADDR_ARGS(flow->dl_src),
2840 in_port->name, vlan);
2842 mac->port.p = in_port;
2843 ofproto_revalidate(br->ofproto, mac_learning_changed(br->ml, mac));
2847 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2848 * dropped. Returns true if they may be forwarded, false if they should be
2851 * If 'have_packet' is true, it indicates that the caller is processing a
2852 * received packet. If 'have_packet' is false, then the caller is just
2853 * revalidating an existing flow because configuration has changed. Either
2854 * way, 'have_packet' only affects logging (there is no point in logging errors
2855 * during revalidation).
2857 * Sets '*in_portp' to the input port. This will be a null pointer if
2858 * flow->in_port does not designate a known input port (in which case
2859 * is_admissible() returns false).
2861 * When returning true, sets '*vlanp' to the effective VLAN of the input
2862 * packet, as returned by flow_get_vlan().
2864 * May also add tags to '*tags', although the current implementation only does
2865 * so in one special case.
2868 is_admissible(struct bridge *br, const struct flow *flow, bool have_packet,
2869 tag_type *tags, int *vlanp, struct port **in_portp)
2871 struct iface *in_iface;
2872 struct port *in_port;
2875 /* Find the interface and port structure for the received packet. */
2876 in_iface = iface_from_dp_ifidx(br, flow->in_port);
2878 /* No interface? Something fishy... */
2880 /* Odd. A few possible reasons here:
2882 * - We deleted an interface but there are still a few packets
2883 * queued up from it.
2885 * - Someone externally added an interface (e.g. with "ovs-dpctl
2886 * add-if") that we don't know about.
2888 * - Packet arrived on the local port but the local port is not
2889 * one of our bridge ports.
2891 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2893 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
2894 "interface %"PRIu16, br->name, flow->in_port);
2900 *in_portp = in_port = in_iface->port;
2901 *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet);
2906 /* Drop frames for reserved multicast addresses. */
2907 if (eth_addr_is_reserved(flow->dl_dst)) {
2911 /* Drop frames on ports reserved for mirroring. */
2912 if (in_port->is_mirror_output_port) {
2914 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2915 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2916 "%s, which is reserved exclusively for mirroring",
2917 br->name, in_port->name);
2922 /* When using LACP, do not accept packets from disabled interfaces. */
2923 if (lacp_negotiated(in_port->lacp) && !in_iface->enabled) {
2927 /* Packets received on non-LACP bonds need special attention to avoid
2929 if (in_port->n_ifaces > 1 && !lacp_negotiated(in_port->lacp)) {
2930 struct mac_entry *mac;
2932 if (eth_addr_is_multicast(flow->dl_dst)) {
2933 *tags |= port_get_active_iface_tag(in_port);
2934 if (in_port->active_iface != in_iface) {
2935 /* Drop all multicast packets on inactive slaves. */
2940 /* Drop all packets for which we have learned a different input
2941 * port, because we probably sent the packet on one slave and got
2942 * it back on the other. Gratuitous ARP packets are an exception
2943 * to this rule: the host has moved to another switch. The exception
2944 * to the exception is if we locked the learning table to avoid
2945 * reflections on bond slaves. If this is the case, just drop the
2947 mac = mac_learning_lookup(br->ml, flow->dl_src, vlan, NULL);
2948 if (mac && mac->port.p != in_port &&
2949 (!is_gratuitous_arp(flow) || mac_entry_is_grat_arp_locked(mac))) {
2957 /* If the composed actions may be applied to any packet in the given 'flow',
2958 * returns true. Otherwise, the actions should only be applied to 'packet', or
2959 * not at all, if 'packet' was NULL. */
2961 process_flow(struct bridge *br, const struct flow *flow,
2962 const struct ofpbuf *packet, struct ofpbuf *actions,
2963 tag_type *tags, uint16_t *nf_output_iface)
2965 struct port *in_port;
2966 struct port *out_port;
2967 struct mac_entry *mac;
2970 /* Check whether we should drop packets in this flow. */
2971 if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) {
2976 /* Learn source MAC (but don't try to learn from revalidation). */
2978 update_learning_table(br, flow, vlan, in_port);
2981 /* Determine output port. */
2982 mac = mac_learning_lookup(br->ml, flow->dl_dst, vlan, tags);
2984 out_port = mac->port.p;
2985 } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) {
2986 /* If we are revalidating but don't have a learning entry then
2987 * eject the flow. Installing a flow that floods packets opens
2988 * up a window of time where we could learn from a packet reflected
2989 * on a bond and blackhole packets before the learning table is
2990 * updated to reflect the correct port. */
2993 out_port = FLOOD_PORT;
2996 /* Don't send packets out their input ports. */
2997 if (in_port == out_port) {
3003 compose_actions(br, flow, vlan, in_port, out_port, tags, actions,
3011 bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
3012 struct ofpbuf *actions, tag_type *tags,
3013 uint16_t *nf_output_iface, void *br_)
3015 struct bridge *br = br_;
3017 COVERAGE_INC(bridge_process_flow);
3018 return process_flow(br, flow, packet, actions, tags, nf_output_iface);
3022 bridge_special_ofhook_cb(const struct flow *flow,
3023 const struct ofpbuf *packet, void *br_)
3025 struct iface *iface;
3026 struct bridge *br = br_;
3028 iface = iface_from_dp_ifidx(br, flow->in_port);
3030 if (flow->dl_type == htons(ETH_TYPE_LACP)) {
3032 if (iface && iface->port->lacp && packet) {
3033 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
3036 COVERAGE_INC(bridge_process_lacp);
3037 lacp_process_pdu(iface->port->lacp, iface, pdu);
3047 bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags,
3048 const struct nlattr *actions,
3050 uint64_t n_bytes, void *br_)
3052 struct bridge *br = br_;
3053 const struct nlattr *a;
3054 struct port *in_port;
3059 /* Feed information from the active flows back into the learning table to
3060 * ensure that table is always in sync with what is actually flowing
3061 * through the datapath.
3063 * We test that 'tags' is nonzero to ensure that only flows that include an
3064 * OFPP_NORMAL action are used for learning. This works because
3065 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
3066 if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) {
3067 update_learning_table(br, flow, vlan, in_port);
3070 /* Account for bond slave utilization. */
3071 if (!br->has_bonded_ports) {
3074 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
3075 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
3076 struct port *out_port = port_from_dp_ifidx(br, nl_attr_get_u32(a));
3077 if (out_port && out_port->n_ifaces >= 2 &&
3078 out_port->bond_mode != BM_AB) {
3079 uint16_t vlan = (flow->vlan_tci
3080 ? vlan_tci_to_vid(flow->vlan_tci)
3082 struct bond_entry *e = lookup_bond_entry(out_port, flow, vlan);
3083 e->tx_bytes += n_bytes;
3090 bridge_account_checkpoint_ofhook_cb(void *br_)
3092 struct bridge *br = br_;
3096 if (!br->has_bonded_ports) {
3101 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3102 if (port->n_ifaces > 1 && port->bond_mode != BM_AB
3103 && now >= port->bond_next_rebalance) {
3104 port->bond_next_rebalance = now + port->bond_rebalance_interval;
3105 bond_rebalance_port(port);
3110 static struct ofhooks bridge_ofhooks = {
3111 bridge_normal_ofhook_cb,
3112 bridge_special_ofhook_cb,
3113 bridge_account_flow_ofhook_cb,
3114 bridge_account_checkpoint_ofhook_cb,
3117 /* Bonding functions. */
3119 /* Statistics for a single interface on a bonded port, used for load-based
3120 * bond rebalancing. */
3121 struct slave_balance {
3122 struct iface *iface; /* The interface. */
3123 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
3125 /* All the "bond_entry"s that are assigned to this interface, in order of
3126 * increasing tx_bytes. */
3127 struct bond_entry **hashes;
3132 bond_mode_to_string(enum bond_mode bm) {
3133 static char *bm_slb = "balance-slb";
3134 static char *bm_ab = "active-backup";
3135 static char *bm_tcp = "balance-tcp";
3138 case BM_SLB: return bm_slb;
3139 case BM_AB: return bm_ab;
3140 case BM_TCP: return bm_tcp;
3147 /* Sorts pointers to pointers to bond_entries in ascending order by the
3148 * interface to which they are assigned, and within a single interface in
3149 * ascending order of bytes transmitted. */
3151 compare_bond_entries(const void *a_, const void *b_)
3153 const struct bond_entry *const *ap = a_;
3154 const struct bond_entry *const *bp = b_;
3155 const struct bond_entry *a = *ap;
3156 const struct bond_entry *b = *bp;
3157 if (a->iface != b->iface) {
3158 return a->iface > b->iface ? 1 : -1;
3159 } else if (a->tx_bytes != b->tx_bytes) {
3160 return a->tx_bytes > b->tx_bytes ? 1 : -1;
3166 /* Sorts slave_balances so that enabled ports come first, and otherwise in
3167 * *descending* order by number of bytes transmitted. */
3169 compare_slave_balance(const void *a_, const void *b_)
3171 const struct slave_balance *a = a_;
3172 const struct slave_balance *b = b_;
3173 if (a->iface->enabled != b->iface->enabled) {
3174 return a->iface->enabled ? -1 : 1;
3175 } else if (a->tx_bytes != b->tx_bytes) {
3176 return a->tx_bytes > b->tx_bytes ? -1 : 1;
3183 swap_bals(struct slave_balance *a, struct slave_balance *b)
3185 struct slave_balance tmp = *a;
3190 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
3191 * given that 'p' (and only 'p') might be in the wrong location.
3193 * This function invalidates 'p', since it might now be in a different memory
3196 resort_bals(struct slave_balance *p,
3197 struct slave_balance bals[], size_t n_bals)
3200 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
3201 swap_bals(p, p - 1);
3203 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
3204 swap_bals(p, p + 1);
3210 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
3212 if (VLOG_IS_DBG_ENABLED()) {
3213 struct ds ds = DS_EMPTY_INITIALIZER;
3214 const struct slave_balance *b;
3216 for (b = bals; b < bals + n_bals; b++) {
3220 ds_put_char(&ds, ',');
3222 ds_put_format(&ds, " %s %"PRIu64"kB",
3223 b->iface->name, b->tx_bytes / 1024);
3225 if (!b->iface->enabled) {
3226 ds_put_cstr(&ds, " (disabled)");
3228 if (b->n_hashes > 0) {
3229 ds_put_cstr(&ds, " (");
3230 for (i = 0; i < b->n_hashes; i++) {
3231 const struct bond_entry *e = b->hashes[i];
3233 ds_put_cstr(&ds, " + ");
3235 ds_put_format(&ds, "h%td: %"PRIu64"kB",
3236 e - port->bond_hash, e->tx_bytes / 1024);
3238 ds_put_cstr(&ds, ")");
3241 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
3246 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
3248 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
3251 struct bond_entry *hash = from->hashes[hash_idx];
3252 struct port *port = from->iface->port;
3253 uint64_t delta = hash->tx_bytes;
3255 assert(port->bond_mode != BM_AB);
3257 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
3258 "from %s to %s (now carrying %"PRIu64"kB and "
3259 "%"PRIu64"kB load, respectively)",
3260 port->name, delta / 1024, hash - port->bond_hash,
3261 from->iface->name, to->iface->name,
3262 (from->tx_bytes - delta) / 1024,
3263 (to->tx_bytes + delta) / 1024);
3265 /* Delete element from from->hashes.
3267 * We don't bother to add the element to to->hashes because not only would
3268 * it require more work, the only purpose it would be to allow that hash to
3269 * be migrated to another slave in this rebalancing run, and there is no
3270 * point in doing that. */
3271 if (hash_idx == 0) {
3274 memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
3275 (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
3279 /* Shift load away from 'from' to 'to'. */
3280 from->tx_bytes -= delta;
3281 to->tx_bytes += delta;
3283 /* Arrange for flows to be revalidated. */
3284 ofproto_revalidate(port->bridge->ofproto, hash->tag);
3285 hash->iface = to->iface;
3286 hash->tag = tag_create_random();
3290 bond_rebalance_port(struct port *port)
3292 struct slave_balance *bals;
3294 struct bond_entry *hashes[BOND_MASK + 1];
3295 struct slave_balance *b, *from, *to;
3296 struct bond_entry *e;
3297 struct iface *iface;
3300 assert(port->bond_mode != BM_AB);
3302 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
3303 * descending order of tx_bytes, so that bals[0] represents the most
3304 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
3307 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
3308 * array for each slave_balance structure, we sort our local array of
3309 * hashes in order by slave, so that all of the hashes for a given slave
3310 * become contiguous in memory, and then we point each 'hashes' members of
3311 * a slave_balance structure to the start of a contiguous group. */
3312 n_bals = port->n_ifaces;
3313 b = bals = xmalloc(n_bals * sizeof *bals);
3314 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3321 assert(b == &bals[n_bals]);
3322 for (i = 0; i <= BOND_MASK; i++) {
3323 hashes[i] = &port->bond_hash[i];
3325 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
3326 for (i = 0; i <= BOND_MASK; i++) {
3332 for (b = bals; b < &bals[n_bals]; b++) {
3333 if (b->iface == e->iface) {
3334 b->tx_bytes += e->tx_bytes;
3336 b->hashes = &hashes[i];
3343 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
3344 log_bals(bals, n_bals, port);
3346 /* Discard slaves that aren't enabled (which were sorted to the back of the
3347 * array earlier). */
3348 while (!bals[n_bals - 1].iface->enabled) {
3355 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
3356 to = &bals[n_bals - 1];
3357 for (from = bals; from < to; ) {
3358 uint64_t overload = from->tx_bytes - to->tx_bytes;
3359 if (overload < to->tx_bytes >> 5 || overload < 100000) {
3360 /* The extra load on 'from' (and all less-loaded slaves), compared
3361 * to that of 'to' (the least-loaded slave), is less than ~3%, or
3362 * it is less than ~1Mbps. No point in rebalancing. */
3364 } else if (from->n_hashes == 1) {
3365 /* 'from' only carries a single MAC hash, so we can't shift any
3366 * load away from it, even though we want to. */
3369 /* 'from' is carrying significantly more load than 'to', and that
3370 * load is split across at least two different hashes. Pick a hash
3371 * to migrate to 'to' (the least-loaded slave), given that doing so
3372 * must decrease the ratio of the load on the two slaves by at
3375 * The sort order we use means that we prefer to shift away the
3376 * smallest hashes instead of the biggest ones. There is little
3377 * reason behind this decision; we could use the opposite sort
3378 * order to shift away big hashes ahead of small ones. */
3381 for (i = 0; i < from->n_hashes; i++) {
3382 double old_ratio, new_ratio;
3383 uint64_t delta = from->hashes[i]->tx_bytes;
3385 if (delta == 0 || from->tx_bytes - delta == 0) {
3386 /* Pointless move. */
3390 order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
3392 if (to->tx_bytes == 0) {
3393 /* Nothing on the new slave, move it. */
3397 old_ratio = (double)from->tx_bytes / to->tx_bytes;
3398 new_ratio = (double)(from->tx_bytes - delta) /
3399 (to->tx_bytes + delta);
3401 if (new_ratio == 0) {
3402 /* Should already be covered but check to prevent division
3407 if (new_ratio < 1) {
3408 new_ratio = 1 / new_ratio;
3411 if (old_ratio - new_ratio > 0.1) {
3412 /* Would decrease the ratio, move it. */
3416 if (i < from->n_hashes) {
3417 bond_shift_load(from, to, i);
3419 /* If the result of the migration changed the relative order of
3420 * 'from' and 'to' swap them back to maintain invariants. */
3421 if (order_swapped) {
3422 swap_bals(from, to);
3425 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
3426 * point to different slave_balance structures. It is only
3427 * valid to do these two operations in a row at all because we
3428 * know that 'from' will not move past 'to' and vice versa. */
3429 resort_bals(from, bals, n_bals);
3430 resort_bals(to, bals, n_bals);
3437 /* Implement exponentially weighted moving average. A weight of 1/2 causes
3438 * historical data to decay to <1% in 7 rebalancing runs. */
3439 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
3451 bond_send_learning_packets(struct port *port)
3453 struct bridge *br = port->bridge;
3454 struct mac_entry *e;
3455 struct ofpbuf packet;
3456 int error, n_packets, n_errors;
3458 if (!port->n_ifaces || !port->active_iface || bond_is_tcp_hash(port)) {
3462 ofpbuf_init(&packet, 128);
3463 error = n_packets = n_errors = 0;
3464 LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
3470 if (e->port.p == port) {
3474 compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
3476 flow_extract(&packet, 0, ODPP_NONE, &flow);
3478 if (!choose_output_iface(port, &flow, e->vlan, &dp_ifidx, &tags)) {
3484 retval = ofproto_send_packet(br->ofproto, dp_ifidx, e->vlan, &packet);
3490 ofpbuf_uninit(&packet);
3493 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3494 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3495 "packets, last error was: %s",
3496 port->name, n_errors, n_packets, strerror(error));
3498 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3499 port->name, n_packets);
3503 /* Bonding unixctl user interface functions. */
3506 bond_unixctl_list(struct unixctl_conn *conn,
3507 const char *args OVS_UNUSED, void *aux OVS_UNUSED)
3509 struct ds ds = DS_EMPTY_INITIALIZER;
3510 const struct bridge *br;
3512 ds_put_cstr(&ds, "bridge\tbond\ttype\tslaves\n");
3514 LIST_FOR_EACH (br, node, &all_bridges) {
3517 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3518 if (port->n_ifaces > 1) {
3519 struct iface *iface;
3521 ds_put_format(&ds, "%s\t%s\t%s\t", br->name, port->name,
3522 bond_mode_to_string(port->bond_mode));
3523 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3524 if (&iface->port_elem != list_front(&port->ifaces)) {
3525 ds_put_cstr(&ds, ", ");
3527 ds_put_cstr(&ds, iface->name);
3529 ds_put_char(&ds, '\n');
3533 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3537 static struct port *
3538 bond_find(const char *name)
3540 const struct bridge *br;
3542 LIST_FOR_EACH (br, node, &all_bridges) {
3545 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
3546 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
3555 bond_unixctl_show(struct unixctl_conn *conn,
3556 const char *args, void *aux OVS_UNUSED)
3558 struct ds ds = DS_EMPTY_INITIALIZER;
3559 const struct port *port;
3560 struct iface *iface;
3562 port = bond_find(args);
3564 unixctl_command_reply(conn, 501, "no such bond");
3568 ds_put_format(&ds, "bond_mode: %s\n",
3569 bond_mode_to_string(port->bond_mode));
3572 ds_put_format(&ds, "lacp: %s\n",
3573 port->lacp_active ? "active" : "passive");
3575 ds_put_cstr(&ds, "lacp: off\n");
3578 if (port->bond_mode != BM_AB) {
3579 ds_put_format(&ds, "bond-hash-algorithm: %s\n",
3580 bond_is_tcp_hash(port) ? "balance-tcp" : "balance-slb");
3584 ds_put_format(&ds, "bond-detect-mode: %s\n",
3585 port->monitor ? "carrier" : "miimon");
3587 if (!port->monitor) {
3588 ds_put_format(&ds, "bond-miimon-interval: %lld\n",
3589 port->miimon_interval);
3592 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
3593 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
3595 if (port->bond_mode != BM_AB) {
3596 ds_put_format(&ds, "next rebalance: %lld ms\n",
3597 port->bond_next_rebalance - time_msec());
3600 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3601 struct bond_entry *be;
3605 ds_put_format(&ds, "\nslave %s: %s\n",
3606 iface->name, iface->enabled ? "enabled" : "disabled");
3607 if (iface == port->active_iface) {
3608 ds_put_cstr(&ds, "\tactive slave\n");
3610 if (iface->delay_expires != LLONG_MAX) {
3611 ds_put_format(&ds, "\t%s expires in %lld ms\n",
3612 iface->enabled ? "downdelay" : "updelay",
3613 iface->delay_expires - time_msec());
3616 if (port->bond_mode == BM_AB) {
3621 memset(&flow, 0, sizeof flow);
3622 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
3623 int hash = be - port->bond_hash;
3624 struct mac_entry *me;
3626 if (be->iface != iface) {
3630 ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n",
3631 hash, be->tx_bytes / 1024);
3633 if (port->bond_mode != BM_SLB) {
3638 LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
3642 memcpy(flow.dl_src, me->mac, ETH_ADDR_LEN);
3643 if (bond_hash_src(me->mac, me->vlan) == hash
3644 && me->port.p != port
3645 && choose_output_iface(port, &flow, me->vlan,
3647 && dp_ifidx == iface->dp_ifidx)
3649 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
3650 ETH_ADDR_ARGS(me->mac));
3655 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3660 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_,
3661 void *aux OVS_UNUSED)
3663 char *args = (char *) args_;
3664 char *save_ptr = NULL;
3665 char *bond_s, *hash_s, *slave_s;
3667 struct iface *iface;
3668 struct bond_entry *entry;
3671 bond_s = strtok_r(args, " ", &save_ptr);
3672 hash_s = strtok_r(NULL, " ", &save_ptr);
3673 slave_s = strtok_r(NULL, " ", &save_ptr);
3675 unixctl_command_reply(conn, 501,
3676 "usage: bond/migrate BOND HASH SLAVE");
3680 port = bond_find(bond_s);
3682 unixctl_command_reply(conn, 501, "no such bond");
3686 if (port->bond_mode != BM_SLB) {
3687 unixctl_command_reply(conn, 501, "not an SLB bond");
3691 if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
3692 hash = atoi(hash_s) & BOND_MASK;
3694 unixctl_command_reply(conn, 501, "bad hash");
3698 iface = port_lookup_iface(port, slave_s);
3700 unixctl_command_reply(conn, 501, "no such slave");
3704 if (!iface->enabled) {
3705 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
3709 entry = &port->bond_hash[hash];
3710 ofproto_revalidate(port->bridge->ofproto, entry->tag);
3711 entry->iface = iface;
3712 entry->tag = tag_create_random();
3713 unixctl_command_reply(conn, 200, "migrated");
3717 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_,
3718 void *aux OVS_UNUSED)
3720 char *args = (char *) args_;
3721 char *save_ptr = NULL;
3722 char *bond_s, *slave_s;
3724 struct iface *iface;
3726 bond_s = strtok_r(args, " ", &save_ptr);
3727 slave_s = strtok_r(NULL, " ", &save_ptr);
3729 unixctl_command_reply(conn, 501,
3730 "usage: bond/set-active-slave BOND SLAVE");
3734 port = bond_find(bond_s);
3736 unixctl_command_reply(conn, 501, "no such bond");
3740 iface = port_lookup_iface(port, slave_s);
3742 unixctl_command_reply(conn, 501, "no such slave");
3746 if (!iface->enabled) {
3747 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
3751 if (port->active_iface != iface) {
3752 ofproto_revalidate(port->bridge->ofproto,
3753 port_get_active_iface_tag(port));
3754 port->active_iface = iface;
3755 VLOG_INFO("port %s: active interface is now %s",
3756 port->name, iface->name);
3757 bond_send_learning_packets(port);
3758 unixctl_command_reply(conn, 200, "done");
3760 unixctl_command_reply(conn, 200, "no change");
3765 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
3767 char *args = (char *) args_;
3768 char *save_ptr = NULL;
3769 char *bond_s, *slave_s;
3771 struct iface *iface;
3773 bond_s = strtok_r(args, " ", &save_ptr);
3774 slave_s = strtok_r(NULL, " ", &save_ptr);
3776 unixctl_command_reply(conn, 501,
3777 "usage: bond/enable/disable-slave BOND SLAVE");
3781 port = bond_find(bond_s);
3783 unixctl_command_reply(conn, 501, "no such bond");
3787 iface = port_lookup_iface(port, slave_s);
3789 unixctl_command_reply(conn, 501, "no such slave");
3793 bond_enable_slave(iface, enable);
3794 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
3798 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args,
3799 void *aux OVS_UNUSED)
3801 enable_slave(conn, args, true);
3805 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args,
3806 void *aux OVS_UNUSED)
3808 enable_slave(conn, args, false);
3812 bond_unixctl_hash(struct unixctl_conn *conn, const char *args_,
3813 void *aux OVS_UNUSED)
3815 char *args = (char *) args_;
3816 uint8_t mac[ETH_ADDR_LEN];
3820 char *mac_s, *vlan_s;
3821 char *save_ptr = NULL;
3823 mac_s = strtok_r(args, " ", &save_ptr);
3824 vlan_s = strtok_r(NULL, " ", &save_ptr);
3827 if (sscanf(vlan_s, "%u", &vlan) != 1) {
3828 unixctl_command_reply(conn, 501, "invalid vlan");
3832 vlan = OFP_VLAN_NONE;
3835 if (sscanf(mac_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
3836 == ETH_ADDR_SCAN_COUNT) {
3837 hash = bond_hash_src(mac, vlan);
3839 hash_cstr = xasprintf("%u", hash);
3840 unixctl_command_reply(conn, 200, hash_cstr);
3843 unixctl_command_reply(conn, 501, "invalid mac");
3850 unixctl_command_register("bond/list", bond_unixctl_list, NULL);
3851 unixctl_command_register("bond/show", bond_unixctl_show, NULL);
3852 unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL);
3853 unixctl_command_register("bond/set-active-slave",
3854 bond_unixctl_set_active_slave, NULL);
3855 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave,
3857 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave,
3859 unixctl_command_register("bond/hash", bond_unixctl_hash, NULL);
3862 /* Port functions. */
3865 lacp_send_pdu_cb(void *aux, const struct lacp_pdu *pdu)
3867 struct iface *iface = aux;
3868 uint8_t ea[ETH_ADDR_LEN];
3871 error = netdev_get_etheraddr(iface->netdev, ea);
3873 struct ofpbuf packet;
3874 struct lacp_pdu *packet_pdu;
3876 ofpbuf_init(&packet, 0);
3877 packet_pdu = compose_packet(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3878 sizeof *packet_pdu);
3879 memcpy(packet_pdu, pdu, sizeof *packet_pdu);
3880 ofproto_send_packet(iface->port->bridge->ofproto,
3881 iface->dp_ifidx, 0, &packet);
3882 ofpbuf_uninit(&packet);
3884 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3885 VLOG_ERR_RL(&rl, "iface %s: failed to obtain Ethernet address "
3886 "(%s)", iface->name, strerror(error));
3891 port_run(struct port *port)
3893 if (port->monitor) {
3896 /* Track carrier going up and down on interfaces. */
3897 while (!netdev_monitor_poll(port->monitor, &devname)) {
3898 struct iface *iface;
3900 iface = port_lookup_iface(port, devname);
3902 iface_update_carrier(iface);
3906 } else if (time_msec() >= port->miimon_next_update) {
3907 struct iface *iface;
3909 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3910 iface_update_carrier(iface);
3912 port->miimon_next_update = time_msec() + port->miimon_interval;
3916 struct iface *iface;
3918 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
3919 lacp_slave_enable(port->lacp, iface, iface->enabled);
3922 lacp_run(port->lacp, lacp_send_pdu_cb);
3929 port_wait(struct port *port)
3931 if (port->monitor) {
3932 netdev_monitor_poll_wait(port->monitor);
3934 poll_timer_wait_until(port->miimon_next_update);
3938 lacp_wait(port->lacp);
3944 static struct port *
3945 port_create(struct bridge *br, const char *name)
3949 port = xzalloc(sizeof *port);
3952 port->trunks = NULL;
3953 port->name = xstrdup(name);
3954 port->active_iface = NULL;
3955 list_init(&port->ifaces);
3957 hmap_insert(&br->ports, &port->hmap_node, hash_string(port->name, 0));
3959 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
3966 get_port_other_config(const struct ovsrec_port *port, const char *key,
3967 const char *default_value)
3971 value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config,
3973 return value ? value : default_value;
3977 get_interface_other_config(const struct ovsrec_interface *iface,
3978 const char *key, const char *default_value)
3982 value = get_ovsrec_key_value(&iface->header_,
3983 &ovsrec_interface_col_other_config, key);
3984 return value ? value : default_value;
3988 port_del_ifaces(struct port *port, const struct ovsrec_port *cfg)
3990 struct iface *iface, *next;
3991 struct sset new_ifaces;
3994 /* Collect list of new interfaces. */
3995 sset_init(&new_ifaces);
3996 for (i = 0; i < cfg->n_interfaces; i++) {
3997 const char *name = cfg->interfaces[i]->name;
3998 sset_add(&new_ifaces, name);
4001 /* Get rid of deleted interfaces. */
4002 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4003 if (!sset_contains(&new_ifaces, iface->name)) {
4004 iface_destroy(iface);
4008 sset_destroy(&new_ifaces);
4011 /* Expires all MAC learning entries associated with 'port' and forces ofproto
4012 * to revalidate every flow. */
4014 port_flush_macs(struct port *port)
4016 struct bridge *br = port->bridge;
4017 struct mac_learning *ml = br->ml;
4018 struct mac_entry *mac, *next_mac;
4021 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
4022 if (mac->port.p == port) {
4023 mac_learning_expire(ml, mac);
4029 port_reconfigure(struct port *port, const struct ovsrec_port *cfg)
4031 const char *detect_mode;
4032 struct sset new_ifaces;
4033 long long int next_rebalance, miimon_next_update, lacp_priority;
4034 bool need_flush = false;
4035 unsigned long *trunks;
4041 /* Update settings. */
4042 port->updelay = cfg->bond_updelay;
4043 if (port->updelay < 0) {
4046 port->downdelay = cfg->bond_downdelay;
4047 if (port->downdelay < 0) {
4048 port->downdelay = 0;
4050 port->bond_rebalance_interval = atoi(
4051 get_port_other_config(cfg, "bond-rebalance-interval", "10000"));
4052 if (port->bond_rebalance_interval < 1000) {
4053 port->bond_rebalance_interval = 1000;
4055 next_rebalance = time_msec() + port->bond_rebalance_interval;
4056 if (port->bond_next_rebalance > next_rebalance) {
4057 port->bond_next_rebalance = next_rebalance;
4060 detect_mode = get_port_other_config(cfg, "bond-detect-mode",
4063 netdev_monitor_destroy(port->monitor);
4064 port->monitor = NULL;
4066 if (strcmp(detect_mode, "miimon")) {
4067 port->monitor = netdev_monitor_create();
4069 if (strcmp(detect_mode, "carrier")) {
4070 VLOG_WARN("port %s: unsupported bond-detect-mode %s, "
4071 "defaulting to carrier", port->name, detect_mode);
4075 port->miimon_interval = atoi(
4076 get_port_other_config(cfg, "bond-miimon-interval", "200"));
4077 if (port->miimon_interval < 100) {
4078 port->miimon_interval = 100;
4080 miimon_next_update = time_msec() + port->miimon_interval;
4081 if (port->miimon_next_update > miimon_next_update) {
4082 port->miimon_next_update = miimon_next_update;
4085 if (!port->cfg->bond_mode ||
4086 !strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_SLB))) {
4087 port->bond_mode = BM_SLB;
4088 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_AB))) {
4089 port->bond_mode = BM_AB;
4090 } else if (!strcmp(port->cfg->bond_mode, bond_mode_to_string(BM_TCP))) {
4091 port->bond_mode = BM_TCP;
4093 port->bond_mode = BM_SLB;
4094 VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s",
4095 port->name, port->cfg->bond_mode,
4096 bond_mode_to_string(port->bond_mode));
4099 /* Add new interfaces and update 'cfg' member of existing ones. */
4100 sset_init(&new_ifaces);
4101 for (i = 0; i < cfg->n_interfaces; i++) {
4102 const struct ovsrec_interface *if_cfg = cfg->interfaces[i];
4103 struct iface *iface;
4105 if (!sset_add(&new_ifaces, if_cfg->name)) {
4106 VLOG_WARN("port %s: %s specified twice as port interface",
4107 port->name, if_cfg->name);
4108 iface_set_ofport(if_cfg, -1);
4112 iface = iface_lookup(port->bridge, if_cfg->name);
4114 if (iface->port != port) {
4115 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
4117 port->bridge->name, if_cfg->name, iface->port->name);
4120 iface->cfg = if_cfg;
4122 iface = iface_create(port, if_cfg);
4125 /* Determine interface type. The local port always has type
4126 * "internal". Other ports take their type from the database and
4127 * default to "system" if none is specified. */
4128 iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal"
4129 : if_cfg->type[0] ? if_cfg->type
4133 atoi(get_interface_other_config(if_cfg, "lacp-port-priority",
4136 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4137 iface->lacp_priority = UINT16_MAX;
4139 iface->lacp_priority = lacp_priority;
4142 sset_destroy(&new_ifaces);
4144 port->lacp_fast = !strcmp(get_port_other_config(cfg, "lacp-time", "slow"),
4148 atoi(get_port_other_config(cfg, "lacp-system-priority", "0"));
4150 if (lacp_priority <= 0 || lacp_priority > UINT16_MAX) {
4151 /* Prefer bondable links if unspecified. */
4152 port->lacp_priority = port->n_ifaces > 1 ? UINT16_MAX - 1 : UINT16_MAX;
4154 port->lacp_priority = lacp_priority;
4157 if (!port->cfg->lacp) {
4158 /* XXX when LACP implementation has been sufficiently tested, enable by
4159 * default and make active on bonded ports. */
4160 lacp_destroy(port->lacp);
4162 } else if (!strcmp(port->cfg->lacp, "off")) {
4163 lacp_destroy(port->lacp);
4165 } else if (!strcmp(port->cfg->lacp, "active")) {
4167 port->lacp = lacp_create();
4169 port->lacp_active = true;
4170 } else if (!strcmp(port->cfg->lacp, "passive")) {
4172 port->lacp = lacp_create();
4174 port->lacp_active = false;
4176 VLOG_WARN("port %s: unknown LACP mode %s",
4177 port->name, port->cfg->lacp);
4178 lacp_destroy(port->lacp);
4185 if (port->n_ifaces < 2) {
4187 if (vlan >= 0 && vlan <= 4095) {
4188 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
4193 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
4194 * they even work as-is. But they have not been tested. */
4195 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
4199 if (port->vlan != vlan) {
4204 /* Get trunked VLANs. */
4206 if (vlan < 0 && cfg->n_trunks) {
4209 trunks = bitmap_allocate(4096);
4211 for (i = 0; i < cfg->n_trunks; i++) {
4212 int trunk = cfg->trunks[i];
4214 bitmap_set1(trunks, trunk);
4220 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
4221 port->name, cfg->n_trunks);
4223 if (n_errors == cfg->n_trunks) {
4224 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
4226 bitmap_free(trunks);
4229 } else if (vlan >= 0 && cfg->n_trunks) {
4230 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
4234 ? port->trunks != NULL
4235 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
4238 bitmap_free(port->trunks);
4239 port->trunks = trunks;
4242 port_flush_macs(port);
4247 port_destroy(struct port *port)
4250 struct bridge *br = port->bridge;
4251 struct iface *iface, *next;
4254 for (i = 0; i < MAX_MIRRORS; i++) {
4255 struct mirror *m = br->mirrors[i];
4256 if (m && m->out_port == port) {
4261 LIST_FOR_EACH_SAFE (iface, next, port_elem, &port->ifaces) {
4262 iface_destroy(iface);
4265 hmap_remove(&br->ports, &port->hmap_node);
4267 VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name);
4269 port_flush_macs(port);
4271 lacp_destroy(port->lacp);
4272 netdev_monitor_destroy(port->monitor);
4273 bitmap_free(port->trunks);
4274 free(port->bond_hash);
4280 static struct port *
4281 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4283 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
4284 return iface ? iface->port : NULL;
4287 static struct port *
4288 port_lookup(const struct bridge *br, const char *name)
4292 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_string(name, 0),
4294 if (!strcmp(port->name, name)) {
4301 static struct iface *
4302 port_lookup_iface(const struct port *port, const char *name)
4304 struct iface *iface = iface_lookup(port->bridge, name);
4305 return iface && iface->port == port ? iface : NULL;
4309 port_update_lacp(struct port *port)
4312 struct iface *iface;
4314 lacp_configure(port->lacp, port->name,
4315 port->bridge->ea, port->lacp_priority,
4316 port->lacp_active, port->lacp_fast);
4318 LIST_FOR_EACH (iface, port_elem, &port->ifaces) {
4319 lacp_slave_register(port->lacp, iface, iface->name,
4320 iface->dp_ifidx, iface->lacp_priority);
4326 port_update_bonding(struct port *port)
4328 if (port->n_ifaces < 2) {
4329 /* Not a bonded port. */
4330 free(port->bond_hash);
4331 port->bond_hash = NULL;
4332 port->bond_fake_iface = false;
4333 port->active_iface = NULL;
4334 port->no_ifaces_tag = 0;
4338 if (port->bond_mode != BM_AB && !port->bond_hash) {
4339 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
4340 for (i = 0; i <= BOND_MASK; i++) {
4341 struct bond_entry *e = &port->bond_hash[i];
4345 port->bond_next_rebalance
4346 = time_msec() + port->bond_rebalance_interval;
4347 } else if (port->bond_mode == BM_AB) {
4348 free(port->bond_hash);
4349 port->bond_hash = NULL;
4352 if (!port->no_ifaces_tag) {
4353 port->no_ifaces_tag = tag_create_random();
4356 if (!port->active_iface) {
4357 bond_choose_active_iface(port);
4360 port->bond_fake_iface = port->cfg->bond_fake_iface;
4361 if (port->bond_fake_iface) {
4362 port->bond_next_fake_iface_update = time_msec();
4368 /* Interface functions. */
4370 static struct iface *
4371 iface_create(struct port *port, const struct ovsrec_interface *if_cfg)
4373 struct bridge *br = port->bridge;
4374 struct iface *iface;
4375 char *name = if_cfg->name;
4377 iface = xzalloc(sizeof *iface);
4379 iface->name = xstrdup(name);
4380 iface->dp_ifidx = -1;
4381 iface->tag = tag_create_random();
4382 iface->delay_expires = LLONG_MAX;
4383 iface->netdev = NULL;
4384 iface->cfg = if_cfg;
4386 shash_add_assert(&br->iface_by_name, iface->name, iface);
4388 list_push_back(&port->ifaces, &iface->port_elem);
4391 if (port->n_ifaces > 1) {
4392 br->has_bonded_ports = true;
4395 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
4403 iface_destroy(struct iface *iface)
4406 struct port *port = iface->port;
4407 struct bridge *br = port->bridge;
4408 bool del_active = port->active_iface == iface;
4410 if (port->bond_hash) {
4411 struct bond_entry *e;
4412 for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
4413 if (e->iface == iface) {
4419 if (iface->port->lacp) {
4420 lacp_slave_unregister(iface->port->lacp, iface);
4423 if (port->monitor && iface->netdev) {
4424 netdev_monitor_remove(port->monitor, iface->netdev);
4427 shash_find_and_delete_assert(&br->iface_by_name, iface->name);
4429 if (iface->dp_ifidx >= 0) {
4430 hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
4433 list_remove(&iface->port_elem);
4436 netdev_close(iface->netdev);
4439 bond_choose_active_iface(port);
4440 bond_send_learning_packets(port);
4446 bridge_flush(port->bridge);
4450 static struct iface *
4451 iface_lookup(const struct bridge *br, const char *name)
4453 return shash_find_data(&br->iface_by_name, name);
4456 static struct iface *
4457 iface_find(const char *name)
4459 const struct bridge *br;
4461 LIST_FOR_EACH (br, node, &all_bridges) {
4462 struct iface *iface = iface_lookup(br, name);
4471 static struct iface *
4472 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
4474 struct iface *iface;
4476 HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
4477 hash_int(dp_ifidx, 0), &br->ifaces) {
4478 if (iface->dp_ifidx == dp_ifidx) {
4485 /* Set Ethernet address of 'iface', if one is specified in the configuration
4488 iface_set_mac(struct iface *iface)
4490 uint8_t ea[ETH_ADDR_LEN];
4492 if (!strcmp(iface->type, "internal")
4493 && iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) {
4494 if (iface->dp_ifidx == ODPP_LOCAL) {
4495 VLOG_ERR("interface %s: ignoring mac in Interface record "
4496 "(use Bridge record to set local port's mac)",
4498 } else if (eth_addr_is_multicast(ea)) {
4499 VLOG_ERR("interface %s: cannot set MAC to multicast address",
4502 int error = netdev_set_etheraddr(iface->netdev, ea);
4504 VLOG_ERR("interface %s: setting MAC failed (%s)",
4505 iface->name, strerror(error));
4511 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
4513 iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport)
4516 ovsrec_interface_set_ofport(if_cfg, &ofport, 1);
4520 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
4522 * The value strings in '*shash' are taken directly from values[], not copied,
4523 * so the caller should not modify or free them. */
4525 shash_from_ovs_idl_map(char **keys, char **values, size_t n,
4526 struct shash *shash)
4531 for (i = 0; i < n; i++) {
4532 shash_add(shash, keys[i], values[i]);
4536 /* Creates 'keys' and 'values' arrays from 'shash'.
4538 * Sets 'keys' and 'values' to heap allocated arrays representing the key-value
4539 * pairs in 'shash'. The caller takes ownership of 'keys' and 'values'. They
4540 * are populated with with strings taken directly from 'shash' and thus have
4541 * the same ownership of the key-value pairs in shash.
4544 shash_to_ovs_idl_map(struct shash *shash,
4545 char ***keys, char ***values, size_t *n)
4549 struct shash_node *sn;
4551 count = shash_count(shash);
4553 k = xmalloc(count * sizeof *k);
4554 v = xmalloc(count * sizeof *v);
4557 SHASH_FOR_EACH(sn, shash) {
4568 struct iface_delete_queues_cbdata {
4569 struct netdev *netdev;
4570 const struct ovsdb_datum *queues;
4574 queue_ids_include(const struct ovsdb_datum *queues, int64_t target)
4576 union ovsdb_atom atom;
4578 atom.integer = target;
4579 return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX;
4583 iface_delete_queues(unsigned int queue_id,
4584 const struct shash *details OVS_UNUSED, void *cbdata_)
4586 struct iface_delete_queues_cbdata *cbdata = cbdata_;
4588 if (!queue_ids_include(cbdata->queues, queue_id)) {
4589 netdev_delete_queue(cbdata->netdev, queue_id);
4594 iface_update_carrier(struct iface *iface)
4596 bool carrier = iface_get_carrier(iface);
4597 if (carrier == iface->up) {
4601 iface->up = carrier;
4602 if (iface->port->lacp) {
4603 lacp_slave_carrier_changed(iface->port->lacp, iface);
4608 iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos)
4610 if (!qos || qos->type[0] == '\0') {
4611 netdev_set_qos(iface->netdev, NULL, NULL);
4613 struct iface_delete_queues_cbdata cbdata;
4614 struct shash details;
4617 /* Configure top-level Qos for 'iface'. */
4618 shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config,
4619 qos->n_other_config, &details);
4620 netdev_set_qos(iface->netdev, qos->type, &details);
4621 shash_destroy(&details);
4623 /* Deconfigure queues that were deleted. */
4624 cbdata.netdev = iface->netdev;
4625 cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER,
4627 netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata);
4629 /* Configure queues for 'iface'. */
4630 for (i = 0; i < qos->n_queues; i++) {
4631 const struct ovsrec_queue *queue = qos->value_queues[i];
4632 unsigned int queue_id = qos->key_queues[i];
4634 shash_from_ovs_idl_map(queue->key_other_config,
4635 queue->value_other_config,
4636 queue->n_other_config, &details);
4637 netdev_set_queue(iface->netdev, queue_id, &details);
4638 shash_destroy(&details);
4644 iface_update_cfm(struct iface *iface)
4648 uint16_t *remote_mps;
4649 struct ovsrec_monitor *mon;
4650 uint8_t maid[CCM_MAID_LEN];
4652 mon = iface->cfg->monitor;
4655 ofproto_iface_clear_cfm(iface->port->bridge->ofproto, iface->dp_ifidx);
4659 if (!cfm_generate_maid(mon->md_name, mon->ma_name, maid)) {
4660 VLOG_WARN("interface %s: Failed to generate MAID.", iface->name);
4664 cfm.mpid = mon->mpid;
4665 cfm.interval = mon->interval ? *mon->interval : 1000;
4667 memcpy(cfm.maid, maid, sizeof cfm.maid);
4669 remote_mps = xzalloc(mon->n_remote_mps * sizeof *remote_mps);
4670 for(i = 0; i < mon->n_remote_mps; i++) {
4671 remote_mps[i] = mon->remote_mps[i]->mpid;
4674 ofproto_iface_set_cfm(iface->port->bridge->ofproto, iface->dp_ifidx,
4675 &cfm, remote_mps, mon->n_remote_mps);
4679 /* Read carrier or miimon status directly from 'iface''s netdev, according to
4680 * how 'iface''s port is configured.
4682 * Returns true if 'iface' is up, false otherwise. */
4684 iface_get_carrier(const struct iface *iface)
4686 return (iface->port->monitor
4687 ? netdev_get_carrier(iface->netdev)
4688 : netdev_get_miimon(iface->netdev));
4691 /* Port mirroring. */
4693 static struct mirror *
4694 mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid)
4698 for (i = 0; i < MAX_MIRRORS; i++) {
4699 struct mirror *m = br->mirrors[i];
4700 if (m && uuid_equals(uuid, &m->uuid)) {
4708 mirror_reconfigure(struct bridge *br)
4710 unsigned long *rspan_vlans;
4714 /* Get rid of deleted mirrors. */
4715 for (i = 0; i < MAX_MIRRORS; i++) {
4716 struct mirror *m = br->mirrors[i];
4718 const struct ovsdb_datum *mc;
4719 union ovsdb_atom atom;
4721 mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID);
4722 atom.uuid = br->mirrors[i]->uuid;
4723 if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) {
4729 /* Add new mirrors and reconfigure existing ones. */
4730 for (i = 0; i < br->cfg->n_mirrors; i++) {
4731 struct ovsrec_mirror *cfg = br->cfg->mirrors[i];
4732 struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid);
4734 mirror_reconfigure_one(m, cfg);
4736 mirror_create(br, cfg);
4740 /* Update port reserved status. */
4741 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4742 port->is_mirror_output_port = false;
4744 for (i = 0; i < MAX_MIRRORS; i++) {
4745 struct mirror *m = br->mirrors[i];
4746 if (m && m->out_port) {
4747 m->out_port->is_mirror_output_port = true;
4751 /* Update flooded vlans (for RSPAN). */
4753 if (br->cfg->n_flood_vlans) {
4754 rspan_vlans = bitmap_allocate(4096);
4756 for (i = 0; i < br->cfg->n_flood_vlans; i++) {
4757 int64_t vlan = br->cfg->flood_vlans[i];
4758 if (vlan >= 0 && vlan < 4096) {
4759 bitmap_set1(rspan_vlans, vlan);
4760 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64,
4763 VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN",
4768 if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) {
4770 mac_learning_flush(br->ml);
4775 mirror_create(struct bridge *br, struct ovsrec_mirror *cfg)
4780 for (i = 0; ; i++) {
4781 if (i >= MAX_MIRRORS) {
4782 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4783 "cannot create %s", br->name, MAX_MIRRORS, cfg->name);
4786 if (!br->mirrors[i]) {
4791 VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name);
4793 mac_learning_flush(br->ml);
4795 br->mirrors[i] = m = xzalloc(sizeof *m);
4796 m->uuid = cfg->header_.uuid;
4799 m->name = xstrdup(cfg->name);
4800 sset_init(&m->src_ports);
4801 sset_init(&m->dst_ports);
4807 mirror_reconfigure_one(m, cfg);
4811 mirror_destroy(struct mirror *m)
4814 struct bridge *br = m->bridge;
4817 HMAP_FOR_EACH (port, hmap_node, &br->ports) {
4818 port->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4819 port->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
4822 sset_destroy(&m->src_ports);
4823 sset_destroy(&m->dst_ports);
4826 m->bridge->mirrors[m->idx] = NULL;
4831 mac_learning_flush(br->ml);
4836 mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports,
4841 for (i = 0; i < n_ports; i++) {
4842 const char *name = ports[i]->name;
4843 if (port_lookup(m->bridge, name)) {
4844 sset_add(names, name);
4846 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4847 "port %s", m->bridge->name, m->name, name);
4853 mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg,
4859 *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan);
4861 for (i = 0; i < cfg->n_select_vlan; i++) {
4862 int64_t vlan = cfg->select_vlan[i];
4863 if (vlan < 0 || vlan > 4095) {
4864 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64,
4865 m->bridge->name, m->name, vlan);
4867 (*vlans)[n_vlans++] = vlan;
4874 vlan_is_mirrored(const struct mirror *m, int vlan)
4878 for (i = 0; i < m->n_vlans; i++) {
4879 if (m->vlans[i] == vlan) {
4887 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
4891 for (i = 0; i < m->n_vlans; i++) {
4892 if (port_trunks_vlan(p, m->vlans[i])) {
4900 mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg)
4902 struct sset src_ports, dst_ports;
4903 mirror_mask_t mirror_bit;
4904 struct port *out_port;
4911 if (strcmp(cfg->name, m->name)) {
4913 m->name = xstrdup(cfg->name);
4916 /* Get output port. */
4917 if (cfg->output_port) {
4918 out_port = port_lookup(m->bridge, cfg->output_port->name);
4920 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4921 m->bridge->name, m->name);
4927 if (cfg->output_vlan) {
4928 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4929 "output vlan; ignoring output vlan",
4930 m->bridge->name, m->name);
4932 } else if (cfg->output_vlan) {
4934 out_vlan = *cfg->output_vlan;
4936 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4937 m->bridge->name, m->name);
4942 sset_init(&src_ports);
4943 sset_init(&dst_ports);
4944 if (cfg->select_all) {
4945 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4946 sset_add(&src_ports, port->name);
4947 sset_add(&dst_ports, port->name);
4952 /* Get ports, and drop duplicates and ports that don't exist. */
4953 mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port,
4955 mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port,
4958 /* Get all the vlans, and drop duplicate and invalid vlans. */
4959 n_vlans = mirror_collect_vlans(m, cfg, &vlans);
4962 /* Update mirror data. */
4963 if (!sset_equals(&m->src_ports, &src_ports)
4964 || !sset_equals(&m->dst_ports, &dst_ports)
4965 || m->n_vlans != n_vlans
4966 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
4967 || m->out_port != out_port
4968 || m->out_vlan != out_vlan) {
4969 bridge_flush(m->bridge);
4970 mac_learning_flush(m->bridge->ml);
4972 sset_swap(&m->src_ports, &src_ports);
4973 sset_swap(&m->dst_ports, &dst_ports);
4976 m->n_vlans = n_vlans;
4977 m->out_port = out_port;
4978 m->out_vlan = out_vlan;
4981 mirror_bit = MIRROR_MASK_C(1) << m->idx;
4982 HMAP_FOR_EACH (port, hmap_node, &m->bridge->ports) {
4983 if (sset_contains(&m->src_ports, port->name)
4986 ? port_trunks_any_mirrored_vlan(m, port)
4987 : vlan_is_mirrored(m, port->vlan)))) {
4988 port->src_mirrors |= mirror_bit;
4990 port->src_mirrors &= ~mirror_bit;
4993 if (sset_contains(&m->dst_ports, port->name)) {
4994 port->dst_mirrors |= mirror_bit;
4996 port->dst_mirrors &= ~mirror_bit;
5001 sset_destroy(&src_ports);
5002 sset_destroy(&dst_ports);